SlideShare ist ein Scribd-Unternehmen logo
1 von 63
Downloaden Sie, um offline zu lesen
Simulation for Autonomous
Driving @Uber ATG
Yu Huang
Sunnyvale,California
Yu.huang07@gmail.com
Reference
• Testing Safety of SDVs by Simulating Perception and Prediction
• LiDARsim: Realistic LiDAR Simulation by Leveraging the Real World
• Recovering and Simulating Pedestrians in the Wild
• S3: Neural Shape, Skeleton, and Skinning Fields for 3D Human Modeling
• SceneGen: Learning to Generate Realistic Traffic Scenes
• TrafficSim: Learning to Simulate Realistic Multi-Agent Behaviors
• GeoSim: Realistic Video Simulation via Geometry-Aware Composition for
Self-Driving
• AdvSim: Generating Safety-Critical Scenarios for Self-Driving Vehicles
• Appendix: (Waymo)
• SurfelGAN: Synthesizing Realistic Sensor Data for Autonomous Driving
Testing the Safety of Self-driving Vehicles by
Simulating Perception and Prediction
• Testing the safety of self driving vehicles in simulation, with an
alternative to sensor simulation.
• Directly simulate the outputs of the self-driving vehicle’s perception
and prediction system, enabling realistic motion planning testing.
• Specifically, use paired data in the form of ground truth labels and
real perception and prediction outputs to train a model that predicts
what the online system will produce.
• Importantly, the inputs to system consists of high definition maps,
bounding boxes, and trajectories, easily sketched in a matter of
minutes, which makes it a much more scalable solution.
Testing the Safety of Self-driving Vehicles by
Simulating Perception and Prediction
The goal is to simulate the outputs of the SDV’s perception and prediction system to realistically test
its motion planner. For each timestep, system ingests an HD map and a set of actors (bounding
boxes and trajectories) and produces noisy outputs similar to those from the real system. To test the
motion planner, mock real outputs with simulated ones.
Testing the Safety of Self-driving Vehicles by
Simulating Perception and Prediction
Perturbation models for perception and prediction simulation. NoNoise assumes perfect perception and
prediction. GaussianNoise and Multimodal Noise use marginal noise distributions to perturb each actor’s
shape, position, and whether it is misdetected. ActorNoise accounts for inter-actor variability by predicting
perturbations conditioned on each actor’s bounding box and positions over time.
Testing the Safety of Self-driving Vehicles by
Simulating Perception and Prediction
ContextNoise for perception and prediction simulation. Given BEV rasterized images of the scene (drawn
from bounding boxes and HD maps), model simulates outputs similar to those from the real perception
and prediction system. It consists of: (i) a shared backbone feature extractor; (ii) a perception head for
simulating bounding box outputs; and (iii) a prediction head for simulating future states outputs.
Testing the Safety of Self-driving Vehicles by
Simulating Perception and Prediction
Simulation results on ATG4D. It
visualizes PLT motion planning
results when given real perception
and prediction (top) versus
simulations from NoNoise (middle)
and ContextNoise (bottom).
ContextNoise faithfully simulates a
misprediction due to multi-modality
and induces a lane-change behavior
from the motion planner.
Testing the Safety of Self-driving Vehicles by
Simulating Perception and Prediction
LiDARsim: Realistic LiDAR Simulation by Leveraging the
RealWorld
• The problem of producing realistic simulations of LiDAR point clouds, the sensor
of preference for most self-driving vehicles.
• By leveraging real data, simulate the complex world more realistically compared
to employing virtual worlds built from CAD/procedural models.
• Towards this goal, first build a large catalog of 3D static maps and 3D dynamic
objects by driving around several cities with our self-driving fleet.
• Then generate scenarios by selecting a scene from catalog and ”virtually” placing
the self-driving vehicle (SDV) and a set of dynamic objects from the catalog in
plausible locations in the scene.
• To produce realistic simulations, develop a simulator that captures both the
power of physics-based and learning-based simulation.
• First utilize ray casting over the 3D scene and then use a deep neural network to
produce deviations from the physics-based simulation, producing realistic LiDAR
point clouds.
LiDARsim: Realistic LiDAR Simulation by Leveraging the
RealWorld
First create the assets from real data, and then compose them into a
scene and simulate the sensor with physics and machine learning.
LiDARsim: Realistic LiDAR Simulation by Leveraging the
RealWorld
Collect real data from multiple trajectories in the same area, remove moving objects, aggregate
and align the data, and create a mesh surfel representation of the background
From left to right: Individual sweep, Accumulated cloud,
Symmetry completion, outlier removal and surfel meshing
LiDARsim: Realistic LiDAR Simulation by Leveraging the
RealWorld
Raydrop physics: Multiple real-world
factors and sensor biases determine if the
signal is detected by LiDAR receiver
Raydrop network: Using ML and real data
to approximate the raydropping process
LiDARsim: Realistic LiDAR Simulation by Leveraging the
RealWorld
Top: Scale of vehicle bank
(displaying several hundred
vehicles out of 25000), Down:
Diversity of vehicle bank colored
by intensity, overlaid on vehicle
dimension scatter plot; Examples
(left to right): opened hood, bikes
on top of vehicle, opened trunk,
pickup with bucket, intensity
shows text, traffic cones on truck,
van with trailer, tractor on truck
LiDARsim: Realistic LiDAR Simulation by Leveraging the
RealWorld
Recovering and Simulating Pedestrians in the Wild
• Sensor simulation is a key component for testing performance of SDVs and
for data augmentation to better train perception systems.
• Recover the shape and motion of pedestrians from sensor readings
captured in the wild by a self-driving car driving around.
• Towards this goal, formulate the problem as energy minimization in a deep
structured model that exploits human shape priors, reprojection
consistency with 2D poses extracted from images, and a ray-caster that
encourages the reconstructed mesh to agree with the LiDAR readings.
• Not require any ground-truth 3D scans or 3D pose annotations.
• Then incorporate the reconstructed pedestrian assets bank in a realistic
LiDAR simulation system by performing motion retargeting;
• The simulated LiDAR data can be used to significantly reduce the amount of
annotated real-world data required for visual perception tasks.
Recovering and Simulating Pedestrians in the Wild
Recover realistic 3D human meshes and poses from sequences of LiDAR and camera readings,
which can then be used in sensor simulation for perception algorithm training and testing.
LiDAR for human Mesh Estimation, (LiME): Given sensory observations, a sensor fusion regression
network predicts the human parameters which minimize the objective function, then perform energy
minimization over the sequence to obtain an optimized shape and 3D pose.
Recovering and Simulating Pedestrians in the Wild
Result on real world data using LiME: (1)
Camera image. (2) Reconstructed mesh. (3)
Ray-casted points on recovered mesh,
overlapped with GT LiDAR points. (4) Side view.
Interpolating a mesh for a query trajectory from
recovered asset mesh sequences.
Recovering and Simulating Pedestrians in the Wild
Quantitative results of method on 3DPW dataset. The sensory input consists of camera
image and the synthetic LiDAR points. Show method using both SMPL model and human
model, and compare with SPIN.
S3: Neural Shape, Skeleton, and Skinning
Fields for 3D Human Modeling
• Constructing and animating humans is an important component for building
virtual worlds in a wide variety of applications such as virtual reality or robotics
testing in simulation.
• As there are exponentially many variations of humans with different shape, pose
and clothing, it is critical to develop methods that can automatically reconstruct
and animate humans at scale from real world data.
• Towards this goal, represent the pedestrian’s shape, pose and skinning weights as
neural implicit functions that are directly learned from data.
• This representation enables us to handle a wide variety of different pedestrian
shapes and poses without explicitly fitting a human parametric body model,
allowing to handle a wider range of human geometries and topologies.
• Generate 3D human animations at scale from a single RGB image (and/or an
optional LiDAR sweep) as input.
S3: Neural Shape, Skeleton, and Skinning
Fields for 3D Human Modeling
Given a single image and/or a single LiDAR sweep as input, model infers shape, skeleton and
skinning jointly, which can then be used to generate animated 3D characters in novel poses.
S3: Neural Shape, Skeleton, and Skinning
Fields for 3D Human Modeling
From left to right: Process input sensor data into spatial feature representations.
Query points adaptively from 3D space and extract their point encoding, used to
query neural implicit representations of shape, pose, and skinning. Apply post-
processing to construct the final explicit representation of an animatable person
S3: Neural Shape, Skeleton, and Skinning
Fields for 3D Human Modeling
SceneGen: Learning to Generate Realistic Traffic Scenes
• For the problem of generating realistic traffic scenes automatically existing
methods typically insert actors into the scene according to a set of hand-crafted
heuristics and are limited in their ability to model the true complexity and
diversity of real traffic scenes, thus inducing a content gap between synthesized
traffic scenes versus real ones.
• As a result, existing simulators (SUMO, CORSIM, VISSIM, and MITSIM) lack the
fidelity necessary to train and test self-driving vehicles.
• To address this limitation, SceneGen—a neural autoregressive model of traffic
scenes that eschews the need for rules and heuristics.
• In particular, given the ego-vehicle state and a high definition map of surrounding
area, SceneGen inserts actors of various classes into the scene and synthesizes
their sizes, orientations, and velocities scenes.
• SceneGen coupled with sensor simulation can be used to train perception models
that generalize to the real world.
SceneGen: Learning to Generate Realistic Traffic Scenes
Autoregressive Traffic Scene Generation Probabilistic Actor Model
Given the ego SDV’s state and an HD map of the surrounding area, SceneGen generates
a traffic scene by inserting actors one at a time. Model each actor probabilistically, as a
product over distributions of its class, position, bounding box, and velocity.
SceneGen: Learning to Generate Realistic Traffic Scenes
The input multi-channel image to SceneGen for ATG4D
SceneGen: Learning to Generate Realistic Traffic Scenes
Traffic scenes generated by SceneGen conditioned on HD maps from ATG4D (top) and Argoverse (bottom).
SceneGen: Learning to Generate Realistic Traffic Scenes
Qualitative comparison of traffic scenes generated by SceneGen and various baselines
SceneGen: Learning to Generate Realistic Traffic Scenes
TrafficSim: Learning to Simulate Realistic Multi-
Agent Behaviors
• To close gap simulation-real world, simulate realistic multi-agent behaviors.
• Heuristic based models that directly encode traffic rules, which cannot capture
irregular maneuvers (nudging, U-turns) and complex interactions (yielding, merging).
• In contrast, leverage real-world data to learn directly from human demonstration
and thus capture a more diverse set of actor behaviors.
• TRAFFICSIM, a multi-agent behavior model for realistic traffic simulation.
• Leverage an implicit latent variable model to parameterize a joint actor policy that
generates socially consistent plans for all actors in the scene jointly.
• To learn a robust policy for long horizon simulation, unroll the policy in training and
optimize through the fully differentiable simulation across time.
• The learning objective incorporates both demonstrations and common sense.
• Exploit trajectories generated by TRAFFICSIM as effective data augmentation for
training better motion planner.
TrafficSim: Learning to Simulate Realistic Multi-
Agent Behaviors
Generating realistic multi-agent behaviors
is a key component for simulation
Complex human driving behavior observed in the real
world: red is actor of interest, green are interacting actors
TrafficSim: Learning to Simulate Realistic Multi-
Agent Behaviors
TRAFFICSIM architecture: global map module (a) is run once per map for repeated simulation
runs. At each timestep, local observation module (b) extracts motion and map features, then
joint behavior module (c) produces a multi-agent plan.
TrafficSim: Learning to Simulate Realistic Multi-
Agent Behaviors
TRAFFICSIM models all actors jointly to simulate realistic traffic scenarios
through time. Sample at each timestep to obtain parallel simulations
TrafficSim: Learning to Simulate Realistic Multi-
Agent Behaviors
Optimize policy with back-propagation through the differentiable simulation (left),
and apply imitation and common sense loss at each simulated state (right).
TrafficSim: Learning to Simulate Realistic Multi-
Agent Behaviors
TrafficSim: Learning to Simulate Realistic Multi-
Agent Behaviors
TrafficSim: Learning to Simulate Realistic Multi-
Agent Behaviors
adaptive weight is a decreasing
function of simulation timestep.
Differentiable relaxation of collision loss
approximates each vehicle as 5 circles and
considers distance between closest centroids.
Time-Adaptive Multi-Task Loss
collision loss
TrafficSim: Learning to Simulate Realistic Multi-
Agent Behaviors
TrafficSim: Learning to Simulate Realistic Multi-
Agent Behaviors
GeoSim: Realistic Video Simulation via Geometry-
Aware Composition for Self-Driving
• GeoSim, a geometry-aware image composition process which synthesizes novel
urban driving scenarios by augmenting existing images with dynamic objects
extracted from other scenes and rendered at novel poses.
• Towards this goal, first build a diverse bank of 3D objects with both realistic
geometry and appearance from sensor data.
• During simulation, perform a novel geometry-aware simulation-by-composition
procedure:
• 1) proposes plausible and realistic object placements into a given scene,
• 2) renders novel views of dynamic objects from the asset bank,
• 3) composes and blends the rendered image segments.
• The resulting synthetic images are realistic, traffic-aware, and geometrically
consistent, allowing it to scale to complex use cases.
• Long-range realistic video simulation across multiple camera sensors, and synthetic
data generation for data augmentation on downstream segmentation tasks.
GeoSim: Realistic Video Simulation via Geometry-
Aware Composition for Self-Driving
Realistic video simulation via geometry-aware composition for self-driving. A data-driven image manipulation
approach that inserts dynamic objects into existing videos. The resulting synthetic video footages are highly
realistic, layout-aware, and geometrically consistent, allowing image simulation to scale to complex use cases.
GeoSim: Realistic Video Simulation via Geometry-
Aware Composition for Self-Driving
Realistic 3D assets creation. Left: multi-view multi-sensor reconstruction network; Right: 3D
asset samples. For each sample we show one of the source images and the 3D mesh.
GeoSim: Realistic Video Simulation via Geometry-
Aware Composition for Self-Driving
3D reconstruction network architecture. Left: Image feature extraction
backbone; Right: Multi-view image fusion block.
GeoSim: Realistic Video Simulation via Geometry-
Aware Composition for Self-Driving
3D-aware object placement, segment retrieval, and temporal simulation.
GeoSim: Realistic Video Simulation via Geometry-
Aware Composition for Self-Driving
Geometry-aware composition with occlusion reasoning followed by an image synthesis module.
GeoSim: Realistic Video Simulation via Geometry-
Aware Composition for Self-Driving
Schematics of shadow generation. (left to right): result without shadow, schematics of
virtual scene, shadow weight (ratio of intensity between rendered image with inserted
object and without inserted object), result with shadow
GeoSim: Realistic Video Simulation via Geometry-
Aware Composition for Self-Driving
GeoSim: Realistic Video Simulation via Geometry-
Aware Composition for Self-Driving
AdvSim: Generating Safety-Critical Scenarios
for Self-Driving Vehicles
• As self-driving systems become better, simulating scenarios where the autonomy
stack may fail becomes more important.
• Those scenarios are generated for a few scenes w.r.t to the planning module that
takes ground-truth actor states as input, which does not scale and cannot identify
all possible autonomy failures, such as perception failures due to occlusion.
• AdvSim, an adversarial framework, generates safety critical scenarios for any LiDAR-
based autonomy system.
• Given an initial traffic scenario, AdvSim modifies the actors’ trajectories in a
physically plausible manner and updates the LiDAR sensor data to match the
perturbed world.
• Importantly, by simulating directly from sensor data, obtain adversarial scenarios
that are safety-critical for the full autonomy stack.
• The robustness and safety of these systems can be further improved by training
them with scenarios generated by AdvSim.
AdvSim: Generating Safety-Critical Scenarios
for Self-Driving Vehicles
The goal is to perturb the maneuvers of interactive actors in an existing scenario with adversarial behaviors that
cause realistic autonomy system failures. Given an existing scenario and its original sensor data, perturb the
scenario and update accordingly how the SDV would observe the LiDAR sensor data based on the new scene
configuration. Then evaluate the autonomy system on the modified scenario, compute an adversarial objective,
and update the proposed perturbation using a search algorithm.
AdvSim: Generating Safety-Critical Scenarios
for Self-Driving Vehicles
AdvSim: Generating Safety-Critical Scenarios
for Self-Driving Vehicles
Realistic LiDAR simulation for scenario perturbations. Given a scenario perturbation on the actors’ motions, the
previously recorded LiDAR data is modified to accurately reflect the updated scene configuration. Remove the
original actor LiDAR observations and replace with simulated actor LiDAR observations at the perturbed
locations, while ensuring sensor realism. The above example perturbs all actors left by 5 meters.
AdvSim: Generating Safety-Critical Scenarios
for Self-Driving Vehicles
AdvSim: Generating Safety-Critical Scenarios
for Self-Driving Vehicles
Visualization of autonomy system’s output plan on original and corresponding adversarial
scenes. A: IL avoids the high-speed lane-changing vehicle behind but collides with the front
one. B: NMP collides with the merging vehicle. C: PLT collides with one vehicle and two
occluded pedestrians at crossroads. D: P3 collides with the crossing pedestrian.
AdvSim: Generating Safety-Critical Scenarios
for Self-Driving Vehicles
AdvSim: Generating Safety-Critical Scenarios
for Self-Driving Vehicles
Appendix (Waymo)
SurfelGAN: Synthesizing Realistic Sensor Data for
Autonomous Driving
• Autonomous driving system development is critically dependent on
the ability to replay complex and diverse traffic scenarios in
simulation.
• In such scenarios, the ability to accurately simulate the vehicle
sensors such as cameras, lidar or radar is essential.
• However, current sensor simulators leverage gaming engines such as
Unreal or Unity, requiring manual creation of environments, objects
and material properties.
• They have limited scalability and fail to produce realistic
approximations of camera, lidar, and radar data without significant
additional work.
SurfelGAN: Synthesizing Realistic Sensor Data for
Autonomous Driving
• SurfelGAN generates realistic scenario sensor data, based only on a
limited amount of lidar and camera data collected by AV.
• It uses texture-mapped surfels to reconstruct the scene from an initial
vehicle pass or set of passes, preserving rich information about object
3D geometry and appearance, as well as the scene conditions.
• Then leverage a SurfelGAN network to reconstruct realistic images for
positions/orientations of the SDV and moving objects in the scene.
• It creates a dataset where two self-driving vehicles observe the same
scene at the same time, which is used to provide additional
evaluation and demonstrate the usefulness of the SurfelGAN model.
SurfelGAN: Synthesizing Realistic Sensor Data for
Autonomous Driving
a) The goal is the generation of camera images for autonomous driving simulation. When
provided with a novel trajectory of the self-driving vehicle in simulation, the system generates
realistic visual sensor data that is useful for downstream modules such as an object detector, a
behavior predictor, or a motion planner. At a high level, the method consists of two steps: b) First,
scan the target environment and reconstruct a scene consisting of rich textured surfels. c) Surfels
are rendered at the camera pose of the novel trajectory, alongside semantic and instance
segmentation masks. Through a GAN, generate realistically looking camera images.
SurfelGAN: Synthesizing Realistic Sensor Data for
Autonomous Driving
SurfelGAN training paradigm. The training setup has two symmetric encoder-decoder generators mapping from surfel
renderings to real images and vice versa. Additionally, there are two discriminators, which specialize in the surfel and the real
domain. The losses are shown as colored arrows. Green: supervised reconstruction loss. Red: adversarial loss. Blue/Yellow:
cycle-consistency losses. When training with paired data, e.g. WOD-TRAIN, the surfel renderings translate to real images,
and apply a one-directional supervised reconstruction loss (SurfelGAN-S) only or add an additional adversarial loss
(SurfelGAN-SA). When training with unpaired data, go either from the surfel renderings (e.g. WOD-TRAIN-NV) or the real
images (e.g. Internal Camera Dataset), use one of the encoder-decoder networks to get to the other domain and back. Then
apply a cycle consistency loss. (SurfelGAN-SAC). The encoder-decoder networks consist of 8 convolutional and 8
deconvolutional layers. Discriminators consist of 5 convolutional layers. All network operate on 256x256 sized input.
SurfelGAN: Synthesizing Realistic Sensor Data for
Autonomous Driving
Qualitative comparison
bewteen different
SurfelGAN variants and
the baseline on WOD-
EVAL under different
weather conditions.
SurfelGAN: Synthesizing Realistic Sensor Data for
Autonomous Driving
Simulation for autonomous driving at uber atg

Weitere ähnliche Inhalte

Was ist angesagt?

Deep VO and SLAM
Deep VO and SLAMDeep VO and SLAM
Deep VO and SLAMYu Huang
 
Driving behaviors for adas and autonomous driving XI
Driving behaviors for adas and autonomous driving XIDriving behaviors for adas and autonomous driving XI
Driving behaviors for adas and autonomous driving XIYu Huang
 
Deep Learning’s Application in Radar Signal Data
Deep Learning’s Application in Radar Signal DataDeep Learning’s Application in Radar Signal Data
Deep Learning’s Application in Radar Signal DataYu Huang
 
BEV Semantic Segmentation
BEV Semantic SegmentationBEV Semantic Segmentation
BEV Semantic SegmentationYu Huang
 
Pedestrian behavior/intention modeling for autonomous driving V
Pedestrian behavior/intention modeling for autonomous driving VPedestrian behavior/intention modeling for autonomous driving V
Pedestrian behavior/intention modeling for autonomous driving VYu Huang
 
3-d interpretation from stereo images for autonomous driving
3-d interpretation from stereo images for autonomous driving3-d interpretation from stereo images for autonomous driving
3-d interpretation from stereo images for autonomous drivingYu Huang
 
Fisheye-Omnidirectional View in Autonomous Driving III
Fisheye-Omnidirectional View in Autonomous Driving IIIFisheye-Omnidirectional View in Autonomous Driving III
Fisheye-Omnidirectional View in Autonomous Driving IIIYu Huang
 
3-d interpretation from single 2-d image for autonomous driving II
3-d interpretation from single 2-d image for autonomous driving II3-d interpretation from single 2-d image for autonomous driving II
3-d interpretation from single 2-d image for autonomous driving IIYu Huang
 
camera-based Lane detection by deep learning
camera-based Lane detection by deep learningcamera-based Lane detection by deep learning
camera-based Lane detection by deep learningYu Huang
 
Driving Behavior for ADAS and Autonomous Driving VIII
Driving Behavior for ADAS and Autonomous Driving VIIIDriving Behavior for ADAS and Autonomous Driving VIII
Driving Behavior for ADAS and Autonomous Driving VIIIYu Huang
 
Camera-based road Lane detection by deep learning III
Camera-based road Lane detection by deep learning IIICamera-based road Lane detection by deep learning III
Camera-based road Lane detection by deep learning IIIYu Huang
 
3-d interpretation from single 2-d image IV
3-d interpretation from single 2-d image IV3-d interpretation from single 2-d image IV
3-d interpretation from single 2-d image IVYu Huang
 
Driving Behavior for ADAS and Autonomous Driving III
Driving Behavior for ADAS and Autonomous Driving IIIDriving Behavior for ADAS and Autonomous Driving III
Driving Behavior for ADAS and Autonomous Driving IIIYu Huang
 
Pedestrian behavior/intention modeling for autonomous driving IV
Pedestrian behavior/intention modeling for autonomous driving IVPedestrian behavior/intention modeling for autonomous driving IV
Pedestrian behavior/intention modeling for autonomous driving IVYu Huang
 
Multi sensor calibration by deep learning
Multi sensor calibration by deep learningMulti sensor calibration by deep learning
Multi sensor calibration by deep learningYu Huang
 
Prediction and planning for self driving at waymo
Prediction and planning for self driving at waymoPrediction and planning for self driving at waymo
Prediction and planning for self driving at waymoYu Huang
 
Annotation tools for ADAS & Autonomous Driving
Annotation tools for ADAS & Autonomous DrivingAnnotation tools for ADAS & Autonomous Driving
Annotation tools for ADAS & Autonomous DrivingYu Huang
 
Jointly mapping, localization, perception, prediction and planning
Jointly mapping, localization, perception, prediction and planningJointly mapping, localization, perception, prediction and planning
Jointly mapping, localization, perception, prediction and planningYu Huang
 
Pedestrian behavior/intention modeling for autonomous driving III
Pedestrian behavior/intention modeling for autonomous driving IIIPedestrian behavior/intention modeling for autonomous driving III
Pedestrian behavior/intention modeling for autonomous driving IIIYu Huang
 

Was ist angesagt? (20)

Deep VO and SLAM
Deep VO and SLAMDeep VO and SLAM
Deep VO and SLAM
 
Driving behaviors for adas and autonomous driving XI
Driving behaviors for adas and autonomous driving XIDriving behaviors for adas and autonomous driving XI
Driving behaviors for adas and autonomous driving XI
 
Deep Learning’s Application in Radar Signal Data
Deep Learning’s Application in Radar Signal DataDeep Learning’s Application in Radar Signal Data
Deep Learning’s Application in Radar Signal Data
 
BEV Semantic Segmentation
BEV Semantic SegmentationBEV Semantic Segmentation
BEV Semantic Segmentation
 
Pedestrian behavior/intention modeling for autonomous driving V
Pedestrian behavior/intention modeling for autonomous driving VPedestrian behavior/intention modeling for autonomous driving V
Pedestrian behavior/intention modeling for autonomous driving V
 
3-d interpretation from stereo images for autonomous driving
3-d interpretation from stereo images for autonomous driving3-d interpretation from stereo images for autonomous driving
3-d interpretation from stereo images for autonomous driving
 
Fisheye-Omnidirectional View in Autonomous Driving III
Fisheye-Omnidirectional View in Autonomous Driving IIIFisheye-Omnidirectional View in Autonomous Driving III
Fisheye-Omnidirectional View in Autonomous Driving III
 
3-d interpretation from single 2-d image for autonomous driving II
3-d interpretation from single 2-d image for autonomous driving II3-d interpretation from single 2-d image for autonomous driving II
3-d interpretation from single 2-d image for autonomous driving II
 
camera-based Lane detection by deep learning
camera-based Lane detection by deep learningcamera-based Lane detection by deep learning
camera-based Lane detection by deep learning
 
Driving Behavior for ADAS and Autonomous Driving VIII
Driving Behavior for ADAS and Autonomous Driving VIIIDriving Behavior for ADAS and Autonomous Driving VIII
Driving Behavior for ADAS and Autonomous Driving VIII
 
Camera-based road Lane detection by deep learning III
Camera-based road Lane detection by deep learning IIICamera-based road Lane detection by deep learning III
Camera-based road Lane detection by deep learning III
 
Survey 1 (project overview)
Survey 1 (project overview)Survey 1 (project overview)
Survey 1 (project overview)
 
3-d interpretation from single 2-d image IV
3-d interpretation from single 2-d image IV3-d interpretation from single 2-d image IV
3-d interpretation from single 2-d image IV
 
Driving Behavior for ADAS and Autonomous Driving III
Driving Behavior for ADAS and Autonomous Driving IIIDriving Behavior for ADAS and Autonomous Driving III
Driving Behavior for ADAS and Autonomous Driving III
 
Pedestrian behavior/intention modeling for autonomous driving IV
Pedestrian behavior/intention modeling for autonomous driving IVPedestrian behavior/intention modeling for autonomous driving IV
Pedestrian behavior/intention modeling for autonomous driving IV
 
Multi sensor calibration by deep learning
Multi sensor calibration by deep learningMulti sensor calibration by deep learning
Multi sensor calibration by deep learning
 
Prediction and planning for self driving at waymo
Prediction and planning for self driving at waymoPrediction and planning for self driving at waymo
Prediction and planning for self driving at waymo
 
Annotation tools for ADAS & Autonomous Driving
Annotation tools for ADAS & Autonomous DrivingAnnotation tools for ADAS & Autonomous Driving
Annotation tools for ADAS & Autonomous Driving
 
Jointly mapping, localization, perception, prediction and planning
Jointly mapping, localization, perception, prediction and planningJointly mapping, localization, perception, prediction and planning
Jointly mapping, localization, perception, prediction and planning
 
Pedestrian behavior/intention modeling for autonomous driving III
Pedestrian behavior/intention modeling for autonomous driving IIIPedestrian behavior/intention modeling for autonomous driving III
Pedestrian behavior/intention modeling for autonomous driving III
 

Ähnlich wie Simulation for autonomous driving at uber atg

10.1109@ICCMC48092.2020.ICCMC-000167.pdf
10.1109@ICCMC48092.2020.ICCMC-000167.pdf10.1109@ICCMC48092.2020.ICCMC-000167.pdf
10.1109@ICCMC48092.2020.ICCMC-000167.pdfmokamojah
 
Fisheye based Perception for Autonomous Driving VI
Fisheye based Perception for Autonomous Driving VIFisheye based Perception for Autonomous Driving VI
Fisheye based Perception for Autonomous Driving VIYu Huang
 
Arindam batabyal literature reviewpresentation
Arindam batabyal literature reviewpresentationArindam batabyal literature reviewpresentation
Arindam batabyal literature reviewpresentationArindam Batabyal
 
"High-resolution 3D Reconstruction on a Mobile Processor," a Presentation fro...
"High-resolution 3D Reconstruction on a Mobile Processor," a Presentation fro..."High-resolution 3D Reconstruction on a Mobile Processor," a Presentation fro...
"High-resolution 3D Reconstruction on a Mobile Processor," a Presentation fro...Edge AI and Vision Alliance
 
fyp presentation of group 43011 final.pptx
fyp presentation of group 43011 final.pptxfyp presentation of group 43011 final.pptx
fyp presentation of group 43011 final.pptxIIEE - NEDUET
 
fusion of Camera and lidar for autonomous driving I
fusion of Camera and lidar for autonomous driving Ifusion of Camera and lidar for autonomous driving I
fusion of Camera and lidar for autonomous driving IYu Huang
 
Presentation Object Recognition And Tracking Project
Presentation Object Recognition And Tracking ProjectPresentation Object Recognition And Tracking Project
Presentation Object Recognition And Tracking ProjectPrathamesh Joshi
 
Driving Behavior for ADAS and Autonomous Driving VII
Driving Behavior for ADAS and Autonomous Driving VIIDriving Behavior for ADAS and Autonomous Driving VII
Driving Behavior for ADAS and Autonomous Driving VIIYu Huang
 
From Sense to Print: Towards Automatic 3D Printing from 3D Sensing Devices
From Sense to Print: Towards Automatic 3D Printing from 3D Sensing DevicesFrom Sense to Print: Towards Automatic 3D Printing from 3D Sensing Devices
From Sense to Print: Towards Automatic 3D Printing from 3D Sensing Devicestoukaigi
 
VR_Module_3_PPT.pptx
VR_Module_3_PPT.pptxVR_Module_3_PPT.pptx
VR_Module_3_PPT.pptxvrfv
 
Understanding the world in 3D with AI.pdf
Understanding the world in 3D with AI.pdfUnderstanding the world in 3D with AI.pdf
Understanding the world in 3D with AI.pdfQualcomm Research
 
3-d interpretation from single 2-d image III
3-d interpretation from single 2-d image III3-d interpretation from single 2-d image III
3-d interpretation from single 2-d image IIIYu Huang
 
Portfolio - Ramsundar K G
Portfolio - Ramsundar K GPortfolio - Ramsundar K G
Portfolio - Ramsundar K GRamsundar K G
 
Experts Vision- Portfolio Jan23
Experts Vision- Portfolio Jan23Experts Vision- Portfolio Jan23
Experts Vision- Portfolio Jan23Hafsa Asad
 
IRJET - Steering Wheel Angle Prediction for Self-Driving Cars
IRJET - Steering Wheel Angle Prediction for Self-Driving CarsIRJET - Steering Wheel Angle Prediction for Self-Driving Cars
IRJET - Steering Wheel Angle Prediction for Self-Driving CarsIRJET Journal
 
Face Detection System on Ada boost Algorithm Using Haar Classifiers
Face Detection System on Ada boost Algorithm Using Haar ClassifiersFace Detection System on Ada boost Algorithm Using Haar Classifiers
Face Detection System on Ada boost Algorithm Using Haar ClassifiersIJMER
 
fusion of Camera and lidar for autonomous driving II
fusion of Camera and lidar for autonomous driving IIfusion of Camera and lidar for autonomous driving II
fusion of Camera and lidar for autonomous driving IIYu Huang
 
VIBE: Video Inference for Human Body Pose and Shape Estimation
VIBE: Video Inference for Human Body Pose and Shape EstimationVIBE: Video Inference for Human Body Pose and Shape Estimation
VIBE: Video Inference for Human Body Pose and Shape EstimationArithmer Inc.
 

Ähnlich wie Simulation for autonomous driving at uber atg (20)

10.1109@ICCMC48092.2020.ICCMC-000167.pdf
10.1109@ICCMC48092.2020.ICCMC-000167.pdf10.1109@ICCMC48092.2020.ICCMC-000167.pdf
10.1109@ICCMC48092.2020.ICCMC-000167.pdf
 
Fisheye based Perception for Autonomous Driving VI
Fisheye based Perception for Autonomous Driving VIFisheye based Perception for Autonomous Driving VI
Fisheye based Perception for Autonomous Driving VI
 
Arindam batabyal literature reviewpresentation
Arindam batabyal literature reviewpresentationArindam batabyal literature reviewpresentation
Arindam batabyal literature reviewpresentation
 
"High-resolution 3D Reconstruction on a Mobile Processor," a Presentation fro...
"High-resolution 3D Reconstruction on a Mobile Processor," a Presentation fro..."High-resolution 3D Reconstruction on a Mobile Processor," a Presentation fro...
"High-resolution 3D Reconstruction on a Mobile Processor," a Presentation fro...
 
fyp presentation of group 43011 final.pptx
fyp presentation of group 43011 final.pptxfyp presentation of group 43011 final.pptx
fyp presentation of group 43011 final.pptx
 
fusion of Camera and lidar for autonomous driving I
fusion of Camera and lidar for autonomous driving Ifusion of Camera and lidar for autonomous driving I
fusion of Camera and lidar for autonomous driving I
 
Presentation Object Recognition And Tracking Project
Presentation Object Recognition And Tracking ProjectPresentation Object Recognition And Tracking Project
Presentation Object Recognition And Tracking Project
 
Driving Behavior for ADAS and Autonomous Driving VII
Driving Behavior for ADAS and Autonomous Driving VIIDriving Behavior for ADAS and Autonomous Driving VII
Driving Behavior for ADAS and Autonomous Driving VII
 
From Sense to Print: Towards Automatic 3D Printing from 3D Sensing Devices
From Sense to Print: Towards Automatic 3D Printing from 3D Sensing DevicesFrom Sense to Print: Towards Automatic 3D Printing from 3D Sensing Devices
From Sense to Print: Towards Automatic 3D Printing from 3D Sensing Devices
 
VR_Module_3_PPT.pptx
VR_Module_3_PPT.pptxVR_Module_3_PPT.pptx
VR_Module_3_PPT.pptx
 
Understanding the world in 3D with AI.pdf
Understanding the world in 3D with AI.pdfUnderstanding the world in 3D with AI.pdf
Understanding the world in 3D with AI.pdf
 
3-d interpretation from single 2-d image III
3-d interpretation from single 2-d image III3-d interpretation from single 2-d image III
3-d interpretation from single 2-d image III
 
Portfolio - Ramsundar K G
Portfolio - Ramsundar K GPortfolio - Ramsundar K G
Portfolio - Ramsundar K G
 
Experts Vision- Portfolio Jan23
Experts Vision- Portfolio Jan23Experts Vision- Portfolio Jan23
Experts Vision- Portfolio Jan23
 
IRJET - Steering Wheel Angle Prediction for Self-Driving Cars
IRJET - Steering Wheel Angle Prediction for Self-Driving CarsIRJET - Steering Wheel Angle Prediction for Self-Driving Cars
IRJET - Steering Wheel Angle Prediction for Self-Driving Cars
 
AR/SLAM for end-users
AR/SLAM for end-usersAR/SLAM for end-users
AR/SLAM for end-users
 
Face Detection System on Ada boost Algorithm Using Haar Classifiers
Face Detection System on Ada boost Algorithm Using Haar ClassifiersFace Detection System on Ada boost Algorithm Using Haar Classifiers
Face Detection System on Ada boost Algorithm Using Haar Classifiers
 
fusion of Camera and lidar for autonomous driving II
fusion of Camera and lidar for autonomous driving IIfusion of Camera and lidar for autonomous driving II
fusion of Camera and lidar for autonomous driving II
 
VIBE: Video Inference for Human Body Pose and Shape Estimation
VIBE: Video Inference for Human Body Pose and Shape EstimationVIBE: Video Inference for Human Body Pose and Shape Estimation
VIBE: Video Inference for Human Body Pose and Shape Estimation
 
ObjectDetection.pptx
ObjectDetection.pptxObjectDetection.pptx
ObjectDetection.pptx
 

Mehr von Yu Huang

Application of Foundation Model for Autonomous Driving
Application of Foundation Model for Autonomous DrivingApplication of Foundation Model for Autonomous Driving
Application of Foundation Model for Autonomous DrivingYu Huang
 
The New Perception Framework in Autonomous Driving: An Introduction of BEV N...
The New Perception Framework  in Autonomous Driving: An Introduction of BEV N...The New Perception Framework  in Autonomous Driving: An Introduction of BEV N...
The New Perception Framework in Autonomous Driving: An Introduction of BEV N...Yu Huang
 
Data Closed Loop in Simulation Test of Autonomous Driving
Data Closed Loop in Simulation Test of Autonomous DrivingData Closed Loop in Simulation Test of Autonomous Driving
Data Closed Loop in Simulation Test of Autonomous DrivingYu Huang
 
Techniques and Challenges in Autonomous Driving
Techniques and Challenges in Autonomous DrivingTechniques and Challenges in Autonomous Driving
Techniques and Challenges in Autonomous DrivingYu Huang
 
BEV Joint Detection and Segmentation
BEV Joint Detection and SegmentationBEV Joint Detection and Segmentation
BEV Joint Detection and SegmentationYu Huang
 
BEV Object Detection and Prediction
BEV Object Detection and PredictionBEV Object Detection and Prediction
BEV Object Detection and PredictionYu Huang
 
Fisheye/Omnidirectional View in Autonomous Driving V
Fisheye/Omnidirectional View in Autonomous Driving VFisheye/Omnidirectional View in Autonomous Driving V
Fisheye/Omnidirectional View in Autonomous Driving VYu Huang
 
Fisheye/Omnidirectional View in Autonomous Driving IV
Fisheye/Omnidirectional View in Autonomous Driving IVFisheye/Omnidirectional View in Autonomous Driving IV
Fisheye/Omnidirectional View in Autonomous Driving IVYu Huang
 
Prediction,Planninng & Control at Baidu
Prediction,Planninng & Control at BaiduPrediction,Planninng & Control at Baidu
Prediction,Planninng & Control at BaiduYu Huang
 
Cruise AI under the Hood
Cruise AI under the HoodCruise AI under the Hood
Cruise AI under the HoodYu Huang
 
LiDAR in the Adverse Weather: Dust, Snow, Rain and Fog (2)
LiDAR in the Adverse Weather: Dust, Snow, Rain and Fog (2)LiDAR in the Adverse Weather: Dust, Snow, Rain and Fog (2)
LiDAR in the Adverse Weather: Dust, Snow, Rain and Fog (2)Yu Huang
 
Scenario-Based Development & Testing for Autonomous Driving
Scenario-Based Development & Testing for Autonomous DrivingScenario-Based Development & Testing for Autonomous Driving
Scenario-Based Development & Testing for Autonomous DrivingYu Huang
 
How to Build a Data Closed-loop Platform for Autonomous Driving?
How to Build a Data Closed-loop Platform for Autonomous Driving?How to Build a Data Closed-loop Platform for Autonomous Driving?
How to Build a Data Closed-loop Platform for Autonomous Driving?Yu Huang
 
Data pipeline and data lake for autonomous driving
Data pipeline and data lake for autonomous drivingData pipeline and data lake for autonomous driving
Data pipeline and data lake for autonomous drivingYu Huang
 
Open Source codes of trajectory prediction & behavior planning
Open Source codes of trajectory prediction & behavior planningOpen Source codes of trajectory prediction & behavior planning
Open Source codes of trajectory prediction & behavior planningYu Huang
 
Lidar in the adverse weather: dust, fog, snow and rain
Lidar in the adverse weather: dust, fog, snow and rainLidar in the adverse weather: dust, fog, snow and rain
Lidar in the adverse weather: dust, fog, snow and rainYu Huang
 
Autonomous Driving of L3/L4 Commercial trucks
Autonomous Driving of L3/L4 Commercial trucksAutonomous Driving of L3/L4 Commercial trucks
Autonomous Driving of L3/L4 Commercial trucksYu Huang
 
3-d interpretation from single 2-d image V
3-d interpretation from single 2-d image V3-d interpretation from single 2-d image V
3-d interpretation from single 2-d image VYu Huang
 
Unsupervised/Self-supervvised visual object tracking
Unsupervised/Self-supervvised visual object trackingUnsupervised/Self-supervvised visual object tracking
Unsupervised/Self-supervvised visual object trackingYu Huang
 
Unsupervised semisupervised semantic or instance segmentation
Unsupervised semisupervised semantic or instance segmentationUnsupervised semisupervised semantic or instance segmentation
Unsupervised semisupervised semantic or instance segmentationYu Huang
 

Mehr von Yu Huang (20)

Application of Foundation Model for Autonomous Driving
Application of Foundation Model for Autonomous DrivingApplication of Foundation Model for Autonomous Driving
Application of Foundation Model for Autonomous Driving
 
The New Perception Framework in Autonomous Driving: An Introduction of BEV N...
The New Perception Framework  in Autonomous Driving: An Introduction of BEV N...The New Perception Framework  in Autonomous Driving: An Introduction of BEV N...
The New Perception Framework in Autonomous Driving: An Introduction of BEV N...
 
Data Closed Loop in Simulation Test of Autonomous Driving
Data Closed Loop in Simulation Test of Autonomous DrivingData Closed Loop in Simulation Test of Autonomous Driving
Data Closed Loop in Simulation Test of Autonomous Driving
 
Techniques and Challenges in Autonomous Driving
Techniques and Challenges in Autonomous DrivingTechniques and Challenges in Autonomous Driving
Techniques and Challenges in Autonomous Driving
 
BEV Joint Detection and Segmentation
BEV Joint Detection and SegmentationBEV Joint Detection and Segmentation
BEV Joint Detection and Segmentation
 
BEV Object Detection and Prediction
BEV Object Detection and PredictionBEV Object Detection and Prediction
BEV Object Detection and Prediction
 
Fisheye/Omnidirectional View in Autonomous Driving V
Fisheye/Omnidirectional View in Autonomous Driving VFisheye/Omnidirectional View in Autonomous Driving V
Fisheye/Omnidirectional View in Autonomous Driving V
 
Fisheye/Omnidirectional View in Autonomous Driving IV
Fisheye/Omnidirectional View in Autonomous Driving IVFisheye/Omnidirectional View in Autonomous Driving IV
Fisheye/Omnidirectional View in Autonomous Driving IV
 
Prediction,Planninng & Control at Baidu
Prediction,Planninng & Control at BaiduPrediction,Planninng & Control at Baidu
Prediction,Planninng & Control at Baidu
 
Cruise AI under the Hood
Cruise AI under the HoodCruise AI under the Hood
Cruise AI under the Hood
 
LiDAR in the Adverse Weather: Dust, Snow, Rain and Fog (2)
LiDAR in the Adverse Weather: Dust, Snow, Rain and Fog (2)LiDAR in the Adverse Weather: Dust, Snow, Rain and Fog (2)
LiDAR in the Adverse Weather: Dust, Snow, Rain and Fog (2)
 
Scenario-Based Development & Testing for Autonomous Driving
Scenario-Based Development & Testing for Autonomous DrivingScenario-Based Development & Testing for Autonomous Driving
Scenario-Based Development & Testing for Autonomous Driving
 
How to Build a Data Closed-loop Platform for Autonomous Driving?
How to Build a Data Closed-loop Platform for Autonomous Driving?How to Build a Data Closed-loop Platform for Autonomous Driving?
How to Build a Data Closed-loop Platform for Autonomous Driving?
 
Data pipeline and data lake for autonomous driving
Data pipeline and data lake for autonomous drivingData pipeline and data lake for autonomous driving
Data pipeline and data lake for autonomous driving
 
Open Source codes of trajectory prediction & behavior planning
Open Source codes of trajectory prediction & behavior planningOpen Source codes of trajectory prediction & behavior planning
Open Source codes of trajectory prediction & behavior planning
 
Lidar in the adverse weather: dust, fog, snow and rain
Lidar in the adverse weather: dust, fog, snow and rainLidar in the adverse weather: dust, fog, snow and rain
Lidar in the adverse weather: dust, fog, snow and rain
 
Autonomous Driving of L3/L4 Commercial trucks
Autonomous Driving of L3/L4 Commercial trucksAutonomous Driving of L3/L4 Commercial trucks
Autonomous Driving of L3/L4 Commercial trucks
 
3-d interpretation from single 2-d image V
3-d interpretation from single 2-d image V3-d interpretation from single 2-d image V
3-d interpretation from single 2-d image V
 
Unsupervised/Self-supervvised visual object tracking
Unsupervised/Self-supervvised visual object trackingUnsupervised/Self-supervvised visual object tracking
Unsupervised/Self-supervvised visual object tracking
 
Unsupervised semisupervised semantic or instance segmentation
Unsupervised semisupervised semantic or instance segmentationUnsupervised semisupervised semantic or instance segmentation
Unsupervised semisupervised semantic or instance segmentation
 

Kürzlich hochgeladen

Thermal Engineering Unit - I & II . ppt
Thermal Engineering  Unit - I & II . pptThermal Engineering  Unit - I & II . ppt
Thermal Engineering Unit - I & II . pptDineshKumar4165
 
Call Girls Walvekar Nagar Call Me 7737669865 Budget Friendly No Advance Booking
Call Girls Walvekar Nagar Call Me 7737669865 Budget Friendly No Advance BookingCall Girls Walvekar Nagar Call Me 7737669865 Budget Friendly No Advance Booking
Call Girls Walvekar Nagar Call Me 7737669865 Budget Friendly No Advance Bookingroncy bisnoi
 
AKTU Computer Networks notes --- Unit 3.pdf
AKTU Computer Networks notes ---  Unit 3.pdfAKTU Computer Networks notes ---  Unit 3.pdf
AKTU Computer Networks notes --- Unit 3.pdfankushspencer015
 
Unit 2- Effective stress & Permeability.pdf
Unit 2- Effective stress & Permeability.pdfUnit 2- Effective stress & Permeability.pdf
Unit 2- Effective stress & Permeability.pdfRagavanV2
 
Double rodded leveling 1 pdf activity 01
Double rodded leveling 1 pdf activity 01Double rodded leveling 1 pdf activity 01
Double rodded leveling 1 pdf activity 01KreezheaRecto
 
Double Revolving field theory-how the rotor develops torque
Double Revolving field theory-how the rotor develops torqueDouble Revolving field theory-how the rotor develops torque
Double Revolving field theory-how the rotor develops torqueBhangaleSonal
 
Top Rated Pune Call Girls Budhwar Peth ⟟ 6297143586 ⟟ Call Me For Genuine Se...
Top Rated  Pune Call Girls Budhwar Peth ⟟ 6297143586 ⟟ Call Me For Genuine Se...Top Rated  Pune Call Girls Budhwar Peth ⟟ 6297143586 ⟟ Call Me For Genuine Se...
Top Rated Pune Call Girls Budhwar Peth ⟟ 6297143586 ⟟ Call Me For Genuine Se...Call Girls in Nagpur High Profile
 
UNIT - IV - Air Compressors and its Performance
UNIT - IV - Air Compressors and its PerformanceUNIT - IV - Air Compressors and its Performance
UNIT - IV - Air Compressors and its Performancesivaprakash250
 
Generative AI or GenAI technology based PPT
Generative AI or GenAI technology based PPTGenerative AI or GenAI technology based PPT
Generative AI or GenAI technology based PPTbhaskargani46
 
Thermal Engineering-R & A / C - unit - V
Thermal Engineering-R & A / C - unit - VThermal Engineering-R & A / C - unit - V
Thermal Engineering-R & A / C - unit - VDineshKumar4165
 
Call Girls Pimpri Chinchwad Call Me 7737669865 Budget Friendly No Advance Boo...
Call Girls Pimpri Chinchwad Call Me 7737669865 Budget Friendly No Advance Boo...Call Girls Pimpri Chinchwad Call Me 7737669865 Budget Friendly No Advance Boo...
Call Girls Pimpri Chinchwad Call Me 7737669865 Budget Friendly No Advance Boo...roncy bisnoi
 
Work-Permit-Receiver-in-Saudi-Aramco.pptx
Work-Permit-Receiver-in-Saudi-Aramco.pptxWork-Permit-Receiver-in-Saudi-Aramco.pptx
Work-Permit-Receiver-in-Saudi-Aramco.pptxJuliansyahHarahap1
 
Top Rated Call Girls In chittoor 📱 {7001035870} VIP Escorts chittoor
Top Rated Call Girls In chittoor 📱 {7001035870} VIP Escorts chittoorTop Rated Call Girls In chittoor 📱 {7001035870} VIP Escorts chittoor
Top Rated Call Girls In chittoor 📱 {7001035870} VIP Escorts chittoordharasingh5698
 
data_management_and _data_science_cheat_sheet.pdf
data_management_and _data_science_cheat_sheet.pdfdata_management_and _data_science_cheat_sheet.pdf
data_management_and _data_science_cheat_sheet.pdfJiananWang21
 
VIP Call Girls Palanpur 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Palanpur 7001035870 Whatsapp Number, 24/07 BookingVIP Call Girls Palanpur 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Palanpur 7001035870 Whatsapp Number, 24/07 Bookingdharasingh5698
 
Unleashing the Power of the SORA AI lastest leap
Unleashing the Power of the SORA AI lastest leapUnleashing the Power of the SORA AI lastest leap
Unleashing the Power of the SORA AI lastest leapRishantSharmaFr
 
notes on Evolution Of Analytic Scalability.ppt
notes on Evolution Of Analytic Scalability.pptnotes on Evolution Of Analytic Scalability.ppt
notes on Evolution Of Analytic Scalability.pptMsecMca
 
Unit 1 - Soil Classification and Compaction.pdf
Unit 1 - Soil Classification and Compaction.pdfUnit 1 - Soil Classification and Compaction.pdf
Unit 1 - Soil Classification and Compaction.pdfRagavanV2
 

Kürzlich hochgeladen (20)

Thermal Engineering Unit - I & II . ppt
Thermal Engineering  Unit - I & II . pptThermal Engineering  Unit - I & II . ppt
Thermal Engineering Unit - I & II . ppt
 
Call Girls Walvekar Nagar Call Me 7737669865 Budget Friendly No Advance Booking
Call Girls Walvekar Nagar Call Me 7737669865 Budget Friendly No Advance BookingCall Girls Walvekar Nagar Call Me 7737669865 Budget Friendly No Advance Booking
Call Girls Walvekar Nagar Call Me 7737669865 Budget Friendly No Advance Booking
 
AKTU Computer Networks notes --- Unit 3.pdf
AKTU Computer Networks notes ---  Unit 3.pdfAKTU Computer Networks notes ---  Unit 3.pdf
AKTU Computer Networks notes --- Unit 3.pdf
 
Unit 2- Effective stress & Permeability.pdf
Unit 2- Effective stress & Permeability.pdfUnit 2- Effective stress & Permeability.pdf
Unit 2- Effective stress & Permeability.pdf
 
Double rodded leveling 1 pdf activity 01
Double rodded leveling 1 pdf activity 01Double rodded leveling 1 pdf activity 01
Double rodded leveling 1 pdf activity 01
 
Double Revolving field theory-how the rotor develops torque
Double Revolving field theory-how the rotor develops torqueDouble Revolving field theory-how the rotor develops torque
Double Revolving field theory-how the rotor develops torque
 
Top Rated Pune Call Girls Budhwar Peth ⟟ 6297143586 ⟟ Call Me For Genuine Se...
Top Rated  Pune Call Girls Budhwar Peth ⟟ 6297143586 ⟟ Call Me For Genuine Se...Top Rated  Pune Call Girls Budhwar Peth ⟟ 6297143586 ⟟ Call Me For Genuine Se...
Top Rated Pune Call Girls Budhwar Peth ⟟ 6297143586 ⟟ Call Me For Genuine Se...
 
UNIT - IV - Air Compressors and its Performance
UNIT - IV - Air Compressors and its PerformanceUNIT - IV - Air Compressors and its Performance
UNIT - IV - Air Compressors and its Performance
 
Generative AI or GenAI technology based PPT
Generative AI or GenAI technology based PPTGenerative AI or GenAI technology based PPT
Generative AI or GenAI technology based PPT
 
Thermal Engineering-R & A / C - unit - V
Thermal Engineering-R & A / C - unit - VThermal Engineering-R & A / C - unit - V
Thermal Engineering-R & A / C - unit - V
 
(INDIRA) Call Girl Bhosari Call Now 8617697112 Bhosari Escorts 24x7
(INDIRA) Call Girl Bhosari Call Now 8617697112 Bhosari Escorts 24x7(INDIRA) Call Girl Bhosari Call Now 8617697112 Bhosari Escorts 24x7
(INDIRA) Call Girl Bhosari Call Now 8617697112 Bhosari Escorts 24x7
 
Call Girls Pimpri Chinchwad Call Me 7737669865 Budget Friendly No Advance Boo...
Call Girls Pimpri Chinchwad Call Me 7737669865 Budget Friendly No Advance Boo...Call Girls Pimpri Chinchwad Call Me 7737669865 Budget Friendly No Advance Boo...
Call Girls Pimpri Chinchwad Call Me 7737669865 Budget Friendly No Advance Boo...
 
Work-Permit-Receiver-in-Saudi-Aramco.pptx
Work-Permit-Receiver-in-Saudi-Aramco.pptxWork-Permit-Receiver-in-Saudi-Aramco.pptx
Work-Permit-Receiver-in-Saudi-Aramco.pptx
 
Top Rated Call Girls In chittoor 📱 {7001035870} VIP Escorts chittoor
Top Rated Call Girls In chittoor 📱 {7001035870} VIP Escorts chittoorTop Rated Call Girls In chittoor 📱 {7001035870} VIP Escorts chittoor
Top Rated Call Girls In chittoor 📱 {7001035870} VIP Escorts chittoor
 
(INDIRA) Call Girl Aurangabad Call Now 8617697112 Aurangabad Escorts 24x7
(INDIRA) Call Girl Aurangabad Call Now 8617697112 Aurangabad Escorts 24x7(INDIRA) Call Girl Aurangabad Call Now 8617697112 Aurangabad Escorts 24x7
(INDIRA) Call Girl Aurangabad Call Now 8617697112 Aurangabad Escorts 24x7
 
data_management_and _data_science_cheat_sheet.pdf
data_management_and _data_science_cheat_sheet.pdfdata_management_and _data_science_cheat_sheet.pdf
data_management_and _data_science_cheat_sheet.pdf
 
VIP Call Girls Palanpur 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Palanpur 7001035870 Whatsapp Number, 24/07 BookingVIP Call Girls Palanpur 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Palanpur 7001035870 Whatsapp Number, 24/07 Booking
 
Unleashing the Power of the SORA AI lastest leap
Unleashing the Power of the SORA AI lastest leapUnleashing the Power of the SORA AI lastest leap
Unleashing the Power of the SORA AI lastest leap
 
notes on Evolution Of Analytic Scalability.ppt
notes on Evolution Of Analytic Scalability.pptnotes on Evolution Of Analytic Scalability.ppt
notes on Evolution Of Analytic Scalability.ppt
 
Unit 1 - Soil Classification and Compaction.pdf
Unit 1 - Soil Classification and Compaction.pdfUnit 1 - Soil Classification and Compaction.pdf
Unit 1 - Soil Classification and Compaction.pdf
 

Simulation for autonomous driving at uber atg

  • 1. Simulation for Autonomous Driving @Uber ATG Yu Huang Sunnyvale,California Yu.huang07@gmail.com
  • 2. Reference • Testing Safety of SDVs by Simulating Perception and Prediction • LiDARsim: Realistic LiDAR Simulation by Leveraging the Real World • Recovering and Simulating Pedestrians in the Wild • S3: Neural Shape, Skeleton, and Skinning Fields for 3D Human Modeling • SceneGen: Learning to Generate Realistic Traffic Scenes • TrafficSim: Learning to Simulate Realistic Multi-Agent Behaviors • GeoSim: Realistic Video Simulation via Geometry-Aware Composition for Self-Driving • AdvSim: Generating Safety-Critical Scenarios for Self-Driving Vehicles • Appendix: (Waymo) • SurfelGAN: Synthesizing Realistic Sensor Data for Autonomous Driving
  • 3. Testing the Safety of Self-driving Vehicles by Simulating Perception and Prediction • Testing the safety of self driving vehicles in simulation, with an alternative to sensor simulation. • Directly simulate the outputs of the self-driving vehicle’s perception and prediction system, enabling realistic motion planning testing. • Specifically, use paired data in the form of ground truth labels and real perception and prediction outputs to train a model that predicts what the online system will produce. • Importantly, the inputs to system consists of high definition maps, bounding boxes, and trajectories, easily sketched in a matter of minutes, which makes it a much more scalable solution.
  • 4. Testing the Safety of Self-driving Vehicles by Simulating Perception and Prediction The goal is to simulate the outputs of the SDV’s perception and prediction system to realistically test its motion planner. For each timestep, system ingests an HD map and a set of actors (bounding boxes and trajectories) and produces noisy outputs similar to those from the real system. To test the motion planner, mock real outputs with simulated ones.
  • 5. Testing the Safety of Self-driving Vehicles by Simulating Perception and Prediction Perturbation models for perception and prediction simulation. NoNoise assumes perfect perception and prediction. GaussianNoise and Multimodal Noise use marginal noise distributions to perturb each actor’s shape, position, and whether it is misdetected. ActorNoise accounts for inter-actor variability by predicting perturbations conditioned on each actor’s bounding box and positions over time.
  • 6. Testing the Safety of Self-driving Vehicles by Simulating Perception and Prediction ContextNoise for perception and prediction simulation. Given BEV rasterized images of the scene (drawn from bounding boxes and HD maps), model simulates outputs similar to those from the real perception and prediction system. It consists of: (i) a shared backbone feature extractor; (ii) a perception head for simulating bounding box outputs; and (iii) a prediction head for simulating future states outputs.
  • 7. Testing the Safety of Self-driving Vehicles by Simulating Perception and Prediction Simulation results on ATG4D. It visualizes PLT motion planning results when given real perception and prediction (top) versus simulations from NoNoise (middle) and ContextNoise (bottom). ContextNoise faithfully simulates a misprediction due to multi-modality and induces a lane-change behavior from the motion planner.
  • 8. Testing the Safety of Self-driving Vehicles by Simulating Perception and Prediction
  • 9. LiDARsim: Realistic LiDAR Simulation by Leveraging the RealWorld • The problem of producing realistic simulations of LiDAR point clouds, the sensor of preference for most self-driving vehicles. • By leveraging real data, simulate the complex world more realistically compared to employing virtual worlds built from CAD/procedural models. • Towards this goal, first build a large catalog of 3D static maps and 3D dynamic objects by driving around several cities with our self-driving fleet. • Then generate scenarios by selecting a scene from catalog and ”virtually” placing the self-driving vehicle (SDV) and a set of dynamic objects from the catalog in plausible locations in the scene. • To produce realistic simulations, develop a simulator that captures both the power of physics-based and learning-based simulation. • First utilize ray casting over the 3D scene and then use a deep neural network to produce deviations from the physics-based simulation, producing realistic LiDAR point clouds.
  • 10. LiDARsim: Realistic LiDAR Simulation by Leveraging the RealWorld First create the assets from real data, and then compose them into a scene and simulate the sensor with physics and machine learning.
  • 11. LiDARsim: Realistic LiDAR Simulation by Leveraging the RealWorld Collect real data from multiple trajectories in the same area, remove moving objects, aggregate and align the data, and create a mesh surfel representation of the background From left to right: Individual sweep, Accumulated cloud, Symmetry completion, outlier removal and surfel meshing
  • 12. LiDARsim: Realistic LiDAR Simulation by Leveraging the RealWorld Raydrop physics: Multiple real-world factors and sensor biases determine if the signal is detected by LiDAR receiver Raydrop network: Using ML and real data to approximate the raydropping process
  • 13. LiDARsim: Realistic LiDAR Simulation by Leveraging the RealWorld Top: Scale of vehicle bank (displaying several hundred vehicles out of 25000), Down: Diversity of vehicle bank colored by intensity, overlaid on vehicle dimension scatter plot; Examples (left to right): opened hood, bikes on top of vehicle, opened trunk, pickup with bucket, intensity shows text, traffic cones on truck, van with trailer, tractor on truck
  • 14. LiDARsim: Realistic LiDAR Simulation by Leveraging the RealWorld
  • 15. Recovering and Simulating Pedestrians in the Wild • Sensor simulation is a key component for testing performance of SDVs and for data augmentation to better train perception systems. • Recover the shape and motion of pedestrians from sensor readings captured in the wild by a self-driving car driving around. • Towards this goal, formulate the problem as energy minimization in a deep structured model that exploits human shape priors, reprojection consistency with 2D poses extracted from images, and a ray-caster that encourages the reconstructed mesh to agree with the LiDAR readings. • Not require any ground-truth 3D scans or 3D pose annotations. • Then incorporate the reconstructed pedestrian assets bank in a realistic LiDAR simulation system by performing motion retargeting; • The simulated LiDAR data can be used to significantly reduce the amount of annotated real-world data required for visual perception tasks.
  • 16. Recovering and Simulating Pedestrians in the Wild Recover realistic 3D human meshes and poses from sequences of LiDAR and camera readings, which can then be used in sensor simulation for perception algorithm training and testing. LiDAR for human Mesh Estimation, (LiME): Given sensory observations, a sensor fusion regression network predicts the human parameters which minimize the objective function, then perform energy minimization over the sequence to obtain an optimized shape and 3D pose.
  • 17. Recovering and Simulating Pedestrians in the Wild Result on real world data using LiME: (1) Camera image. (2) Reconstructed mesh. (3) Ray-casted points on recovered mesh, overlapped with GT LiDAR points. (4) Side view. Interpolating a mesh for a query trajectory from recovered asset mesh sequences.
  • 18. Recovering and Simulating Pedestrians in the Wild Quantitative results of method on 3DPW dataset. The sensory input consists of camera image and the synthetic LiDAR points. Show method using both SMPL model and human model, and compare with SPIN.
  • 19. S3: Neural Shape, Skeleton, and Skinning Fields for 3D Human Modeling • Constructing and animating humans is an important component for building virtual worlds in a wide variety of applications such as virtual reality or robotics testing in simulation. • As there are exponentially many variations of humans with different shape, pose and clothing, it is critical to develop methods that can automatically reconstruct and animate humans at scale from real world data. • Towards this goal, represent the pedestrian’s shape, pose and skinning weights as neural implicit functions that are directly learned from data. • This representation enables us to handle a wide variety of different pedestrian shapes and poses without explicitly fitting a human parametric body model, allowing to handle a wider range of human geometries and topologies. • Generate 3D human animations at scale from a single RGB image (and/or an optional LiDAR sweep) as input.
  • 20. S3: Neural Shape, Skeleton, and Skinning Fields for 3D Human Modeling Given a single image and/or a single LiDAR sweep as input, model infers shape, skeleton and skinning jointly, which can then be used to generate animated 3D characters in novel poses.
  • 21. S3: Neural Shape, Skeleton, and Skinning Fields for 3D Human Modeling From left to right: Process input sensor data into spatial feature representations. Query points adaptively from 3D space and extract their point encoding, used to query neural implicit representations of shape, pose, and skinning. Apply post- processing to construct the final explicit representation of an animatable person
  • 22. S3: Neural Shape, Skeleton, and Skinning Fields for 3D Human Modeling
  • 23. SceneGen: Learning to Generate Realistic Traffic Scenes • For the problem of generating realistic traffic scenes automatically existing methods typically insert actors into the scene according to a set of hand-crafted heuristics and are limited in their ability to model the true complexity and diversity of real traffic scenes, thus inducing a content gap between synthesized traffic scenes versus real ones. • As a result, existing simulators (SUMO, CORSIM, VISSIM, and MITSIM) lack the fidelity necessary to train and test self-driving vehicles. • To address this limitation, SceneGen—a neural autoregressive model of traffic scenes that eschews the need for rules and heuristics. • In particular, given the ego-vehicle state and a high definition map of surrounding area, SceneGen inserts actors of various classes into the scene and synthesizes their sizes, orientations, and velocities scenes. • SceneGen coupled with sensor simulation can be used to train perception models that generalize to the real world.
  • 24. SceneGen: Learning to Generate Realistic Traffic Scenes Autoregressive Traffic Scene Generation Probabilistic Actor Model Given the ego SDV’s state and an HD map of the surrounding area, SceneGen generates a traffic scene by inserting actors one at a time. Model each actor probabilistically, as a product over distributions of its class, position, bounding box, and velocity.
  • 25. SceneGen: Learning to Generate Realistic Traffic Scenes The input multi-channel image to SceneGen for ATG4D
  • 26. SceneGen: Learning to Generate Realistic Traffic Scenes Traffic scenes generated by SceneGen conditioned on HD maps from ATG4D (top) and Argoverse (bottom).
  • 27. SceneGen: Learning to Generate Realistic Traffic Scenes Qualitative comparison of traffic scenes generated by SceneGen and various baselines
  • 28. SceneGen: Learning to Generate Realistic Traffic Scenes
  • 29. TrafficSim: Learning to Simulate Realistic Multi- Agent Behaviors • To close gap simulation-real world, simulate realistic multi-agent behaviors. • Heuristic based models that directly encode traffic rules, which cannot capture irregular maneuvers (nudging, U-turns) and complex interactions (yielding, merging). • In contrast, leverage real-world data to learn directly from human demonstration and thus capture a more diverse set of actor behaviors. • TRAFFICSIM, a multi-agent behavior model for realistic traffic simulation. • Leverage an implicit latent variable model to parameterize a joint actor policy that generates socially consistent plans for all actors in the scene jointly. • To learn a robust policy for long horizon simulation, unroll the policy in training and optimize through the fully differentiable simulation across time. • The learning objective incorporates both demonstrations and common sense. • Exploit trajectories generated by TRAFFICSIM as effective data augmentation for training better motion planner.
  • 30. TrafficSim: Learning to Simulate Realistic Multi- Agent Behaviors Generating realistic multi-agent behaviors is a key component for simulation Complex human driving behavior observed in the real world: red is actor of interest, green are interacting actors
  • 31. TrafficSim: Learning to Simulate Realistic Multi- Agent Behaviors TRAFFICSIM architecture: global map module (a) is run once per map for repeated simulation runs. At each timestep, local observation module (b) extracts motion and map features, then joint behavior module (c) produces a multi-agent plan.
  • 32. TrafficSim: Learning to Simulate Realistic Multi- Agent Behaviors TRAFFICSIM models all actors jointly to simulate realistic traffic scenarios through time. Sample at each timestep to obtain parallel simulations
  • 33. TrafficSim: Learning to Simulate Realistic Multi- Agent Behaviors Optimize policy with back-propagation through the differentiable simulation (left), and apply imitation and common sense loss at each simulated state (right).
  • 34. TrafficSim: Learning to Simulate Realistic Multi- Agent Behaviors
  • 35. TrafficSim: Learning to Simulate Realistic Multi- Agent Behaviors
  • 36. TrafficSim: Learning to Simulate Realistic Multi- Agent Behaviors adaptive weight is a decreasing function of simulation timestep. Differentiable relaxation of collision loss approximates each vehicle as 5 circles and considers distance between closest centroids. Time-Adaptive Multi-Task Loss collision loss
  • 37. TrafficSim: Learning to Simulate Realistic Multi- Agent Behaviors
  • 38. TrafficSim: Learning to Simulate Realistic Multi- Agent Behaviors
  • 39. GeoSim: Realistic Video Simulation via Geometry- Aware Composition for Self-Driving • GeoSim, a geometry-aware image composition process which synthesizes novel urban driving scenarios by augmenting existing images with dynamic objects extracted from other scenes and rendered at novel poses. • Towards this goal, first build a diverse bank of 3D objects with both realistic geometry and appearance from sensor data. • During simulation, perform a novel geometry-aware simulation-by-composition procedure: • 1) proposes plausible and realistic object placements into a given scene, • 2) renders novel views of dynamic objects from the asset bank, • 3) composes and blends the rendered image segments. • The resulting synthetic images are realistic, traffic-aware, and geometrically consistent, allowing it to scale to complex use cases. • Long-range realistic video simulation across multiple camera sensors, and synthetic data generation for data augmentation on downstream segmentation tasks.
  • 40. GeoSim: Realistic Video Simulation via Geometry- Aware Composition for Self-Driving Realistic video simulation via geometry-aware composition for self-driving. A data-driven image manipulation approach that inserts dynamic objects into existing videos. The resulting synthetic video footages are highly realistic, layout-aware, and geometrically consistent, allowing image simulation to scale to complex use cases.
  • 41. GeoSim: Realistic Video Simulation via Geometry- Aware Composition for Self-Driving Realistic 3D assets creation. Left: multi-view multi-sensor reconstruction network; Right: 3D asset samples. For each sample we show one of the source images and the 3D mesh.
  • 42. GeoSim: Realistic Video Simulation via Geometry- Aware Composition for Self-Driving 3D reconstruction network architecture. Left: Image feature extraction backbone; Right: Multi-view image fusion block.
  • 43. GeoSim: Realistic Video Simulation via Geometry- Aware Composition for Self-Driving 3D-aware object placement, segment retrieval, and temporal simulation.
  • 44. GeoSim: Realistic Video Simulation via Geometry- Aware Composition for Self-Driving Geometry-aware composition with occlusion reasoning followed by an image synthesis module.
  • 45. GeoSim: Realistic Video Simulation via Geometry- Aware Composition for Self-Driving Schematics of shadow generation. (left to right): result without shadow, schematics of virtual scene, shadow weight (ratio of intensity between rendered image with inserted object and without inserted object), result with shadow
  • 46. GeoSim: Realistic Video Simulation via Geometry- Aware Composition for Self-Driving
  • 47. GeoSim: Realistic Video Simulation via Geometry- Aware Composition for Self-Driving
  • 48. AdvSim: Generating Safety-Critical Scenarios for Self-Driving Vehicles • As self-driving systems become better, simulating scenarios where the autonomy stack may fail becomes more important. • Those scenarios are generated for a few scenes w.r.t to the planning module that takes ground-truth actor states as input, which does not scale and cannot identify all possible autonomy failures, such as perception failures due to occlusion. • AdvSim, an adversarial framework, generates safety critical scenarios for any LiDAR- based autonomy system. • Given an initial traffic scenario, AdvSim modifies the actors’ trajectories in a physically plausible manner and updates the LiDAR sensor data to match the perturbed world. • Importantly, by simulating directly from sensor data, obtain adversarial scenarios that are safety-critical for the full autonomy stack. • The robustness and safety of these systems can be further improved by training them with scenarios generated by AdvSim.
  • 49. AdvSim: Generating Safety-Critical Scenarios for Self-Driving Vehicles The goal is to perturb the maneuvers of interactive actors in an existing scenario with adversarial behaviors that cause realistic autonomy system failures. Given an existing scenario and its original sensor data, perturb the scenario and update accordingly how the SDV would observe the LiDAR sensor data based on the new scene configuration. Then evaluate the autonomy system on the modified scenario, compute an adversarial objective, and update the proposed perturbation using a search algorithm.
  • 50. AdvSim: Generating Safety-Critical Scenarios for Self-Driving Vehicles
  • 51. AdvSim: Generating Safety-Critical Scenarios for Self-Driving Vehicles Realistic LiDAR simulation for scenario perturbations. Given a scenario perturbation on the actors’ motions, the previously recorded LiDAR data is modified to accurately reflect the updated scene configuration. Remove the original actor LiDAR observations and replace with simulated actor LiDAR observations at the perturbed locations, while ensuring sensor realism. The above example perturbs all actors left by 5 meters.
  • 52. AdvSim: Generating Safety-Critical Scenarios for Self-Driving Vehicles
  • 53. AdvSim: Generating Safety-Critical Scenarios for Self-Driving Vehicles Visualization of autonomy system’s output plan on original and corresponding adversarial scenes. A: IL avoids the high-speed lane-changing vehicle behind but collides with the front one. B: NMP collides with the merging vehicle. C: PLT collides with one vehicle and two occluded pedestrians at crossroads. D: P3 collides with the crossing pedestrian.
  • 54. AdvSim: Generating Safety-Critical Scenarios for Self-Driving Vehicles
  • 55. AdvSim: Generating Safety-Critical Scenarios for Self-Driving Vehicles
  • 57. SurfelGAN: Synthesizing Realistic Sensor Data for Autonomous Driving • Autonomous driving system development is critically dependent on the ability to replay complex and diverse traffic scenarios in simulation. • In such scenarios, the ability to accurately simulate the vehicle sensors such as cameras, lidar or radar is essential. • However, current sensor simulators leverage gaming engines such as Unreal or Unity, requiring manual creation of environments, objects and material properties. • They have limited scalability and fail to produce realistic approximations of camera, lidar, and radar data without significant additional work.
  • 58. SurfelGAN: Synthesizing Realistic Sensor Data for Autonomous Driving • SurfelGAN generates realistic scenario sensor data, based only on a limited amount of lidar and camera data collected by AV. • It uses texture-mapped surfels to reconstruct the scene from an initial vehicle pass or set of passes, preserving rich information about object 3D geometry and appearance, as well as the scene conditions. • Then leverage a SurfelGAN network to reconstruct realistic images for positions/orientations of the SDV and moving objects in the scene. • It creates a dataset where two self-driving vehicles observe the same scene at the same time, which is used to provide additional evaluation and demonstrate the usefulness of the SurfelGAN model.
  • 59. SurfelGAN: Synthesizing Realistic Sensor Data for Autonomous Driving a) The goal is the generation of camera images for autonomous driving simulation. When provided with a novel trajectory of the self-driving vehicle in simulation, the system generates realistic visual sensor data that is useful for downstream modules such as an object detector, a behavior predictor, or a motion planner. At a high level, the method consists of two steps: b) First, scan the target environment and reconstruct a scene consisting of rich textured surfels. c) Surfels are rendered at the camera pose of the novel trajectory, alongside semantic and instance segmentation masks. Through a GAN, generate realistically looking camera images.
  • 60. SurfelGAN: Synthesizing Realistic Sensor Data for Autonomous Driving SurfelGAN training paradigm. The training setup has two symmetric encoder-decoder generators mapping from surfel renderings to real images and vice versa. Additionally, there are two discriminators, which specialize in the surfel and the real domain. The losses are shown as colored arrows. Green: supervised reconstruction loss. Red: adversarial loss. Blue/Yellow: cycle-consistency losses. When training with paired data, e.g. WOD-TRAIN, the surfel renderings translate to real images, and apply a one-directional supervised reconstruction loss (SurfelGAN-S) only or add an additional adversarial loss (SurfelGAN-SA). When training with unpaired data, go either from the surfel renderings (e.g. WOD-TRAIN-NV) or the real images (e.g. Internal Camera Dataset), use one of the encoder-decoder networks to get to the other domain and back. Then apply a cycle consistency loss. (SurfelGAN-SAC). The encoder-decoder networks consist of 8 convolutional and 8 deconvolutional layers. Discriminators consist of 5 convolutional layers. All network operate on 256x256 sized input.
  • 61. SurfelGAN: Synthesizing Realistic Sensor Data for Autonomous Driving Qualitative comparison bewteen different SurfelGAN variants and the baseline on WOD- EVAL under different weather conditions.
  • 62. SurfelGAN: Synthesizing Realistic Sensor Data for Autonomous Driving