Processing & Properties of Floor and Wall Tiles.pptx
Prediction,Planninng & Control at Baidu
1. Data Driven Method of Prediction
and Planning for Autonomous
Driving at Baidu Apollo
Yu Huang
Sunnyvale, California
Yu.huang07@gmail.com
2. References
• An Auto-tuning Framework for Autonomous Vehicles
• Data Driven Prediction Architecture for Autonomous Driving and its
Application on Apollo Platform
• DL-IAPS and PJSO: A Path/Speed Decoupled Trajectory Optimization
and its Application in Autonomous Driving
• TDR-OBCA: A Reliable Planner for Autonomous Driving in Free-Space
Environment
• DRF: A Framework for High-Accuracy Autonomous Driving Vehicle
Modeling
• A Learning-Based Tune-Free Control Framework for Large Scale
Autonomous Driving System Deployment
3. An Auto-tuning Framework for Autonomous Vehicles
• Many autonomous driving motion planners generate trajectories by optimizing a reward/cost
functional.
• Designing and tuning a high-performance reward/cost functional for Level-4 autonomous driving
vehicles is challenging.
• Traditionally, reward/cost functional tuning involves substantial human effort and time spent on
both simulations and road tests.
• As the scenario becomes more complicated, tuning to improve the motion planner performance
becomes increasingly difficult.
• A data-driven auto-tuning framework based on the Apollo autonomous driving framework.
• The framework includes a rank-based conditional inverse reinforcement learning algorithm, an
offline training strategy and an automatic method of collecting and labeling data.
• The advantages: First, compared to that of most inverse reinforcement learning algorithms, the
algorithm training is efficient and capable of being applied to different scenarios; Second, the
offline training strategy offers a safe way to adjust the parameters before public road testing;
Third, the expert driving data and information about the surrounding environment are collected
and automatically labeled, which considerably reduces the manual effort.
7. An Auto-tuning Framework for Autonomous Vehicles
One frame of the optimal trajectory with respect to the tuned
reward functional compared to the human driving trajectory.
8. Data Driven Prediction Architecture for Autonomous
Driving and its Application on Apollo Platform
• Autonomous Driving vehicles (ADV) are on road with large scales.
• For safe and efficient operations, ADVs must be able to predict the future states
and iterative with road entities in complex, real-world driving scenarios.
• How to migrate a well-trained prediction model from one geo-fenced area to
another is essential in scaling the ADV operation and is difficult most of the time
since the terrains, traffic rules, entities distributions, driving/walking patterns
would be largely different in different geo-fenced operation areas.
• In this paper, a highly automated learning-based prediction model pipeline, which
has been deployed on Baidu Apollo self-driving platform, to support different
prediction learning sub-modules’ data annotation, feature extraction, model
training/tuning and deployment.
• This pipeline is completely automatic without any human intervention and shows
an up to 400% efficiency increase in parameter tuning, when deployed at scale in
different scenarios across nations.
9. Data Driven Prediction Architecture for Autonomous
Driving and its Application on Apollo Platform
Onboard architecture of
the prediction module on
Apollo autonomous driving
open-source platform
10. Data Driven Prediction Architecture for Autonomous
Driving and its Application on Apollo Platform
Offboard architecture of
the prediction module on
Apollo autonomous driving
open-source platform
11. Data Driven Prediction Architecture for Autonomous
Driving and its Application on Apollo Platform
Structure of semantic map + LSTM model
12. Data Driven Prediction Architecture for Autonomous
Driving and its Application on Apollo Platform
Workflow of intention prediction
and post trajectory generation.
13. Data Driven Prediction Architecture for Autonomous
Driving and its Application on Apollo Platform
14. DL-IAPS and PJSO: A Path/Speed Decoupled Trajectory
Optimization and its Application in Autonomous Driving
• This is a free space trajectory optimization algorithm of autonomous
driving vehicle, which decouples the collision-free trajectory planning
problem into a Dual-Loop Iterative Anchoring Path Smoothing (DL-
IAPS) and a Piece-wise Jerk Speed Optimization (PJSO).
• The work leads to remarkable driving performance improvements
including more precise collision avoidance, higher control feasibility
and better driving comfort, as those are often hard to realize in other
existing path/speed decoupled trajectory optimization methods.
15. Open Space Planner Architecture
DL-IAPS and PJSO: A Path/Speed Decoupled Trajectory
Optimization and its Application in Autonomous Driving
16. DL-IAPS and PJSO: A Path/Speed Decoupled Trajectory
Optimization and its Application in Autonomous Driving
17. On-road test vehicle controller architecture
DL-IAPS and PJSO: A Path/Speed Decoupled Trajectory
Optimization and its Application in Autonomous Driving
18. DL-IAPS and PJSO: A Path/Speed Decoupled Trajectory
Optimization and its Application in Autonomous Driving
19. TDR-OBCA: A Reliable Planner for Autonomous
Driving in Free-Space Environment
• This is an optimization-based collision avoidance trajectory generation
method for autonomous driving in free-space environments, with
enhanced robustness, driving comfort and efficiency.
• Starting from the hybrid optimization-based framework, here comes
two warm start methods, temporal and dual variable warm starts, to
improve the efficiency.
TDR-OBCA Architecture
20. TDR-OBCA: A Reliable Planner for Autonomous
Driving in Free-Space Environment
TDR-OBCA application in Apollo
Autonomous Driving Platform
22. DRF: A Framework for High-Accuracy
Autonomous Driving Vehicle Modeling
• An accurate vehicle dynamic model is the key to bridge the gap between
simulation and real road test in autonomous driving.
• This is a Dynamic model-Residual correction model Framework (DRF) for
vehicle dynamic modeling (DM).
• On top of any existing open-loop dynamic model, this framework builds a
Residual Correction Model (RCM) by integrating deep NN with Sparse
Variational Gaussian Process (SVGP) model.
• RCM takes a sequence of vehicle control commands and dynamic status for
a certain time duration as modeling inputs, extracts underlying context
from this sequence with deep encoder networks, and predicts open-loop
dynamic model prediction errors.
• Five vehicle dynamic models are derived from DRF via encoder variation.
23. DRF: A Framework for High-Accuracy
Autonomous Driving Vehicle Modeling
Apollo DRF Machine Learning Platform
Dynamic residual correction Framework (DRF) Structure
24. DRF: A Framework for High-Accuracy
Autonomous Driving Vehicle Modeling
Residual correction module (RCM) structure with transformer encoder
Residual correction module (RCM) structure with CNN encoder
25. DRF: A Framework for High-Accuracy
Autonomous Driving Vehicle Modeling
26. DRF: A Framework for High-Accuracy
Autonomous Driving Vehicle Modeling
Control-in-the-Loop Simulation
27. DRF: A Framework for High-Accuracy
Autonomous Driving Vehicle Modeling
28. A Learning-Based Tune-Free Control Framework for
Large Scale Autonomous Driving System Deployment
• This is the design of a tune-free (human-out-of-the-loop parameter tuning)
control framework, aiming at accelerating large scale autonomous driving
system deployed on various vehicles and driving environments.
• The framework consists of three ML-based procedures, which jointly
automate the control parameter tuning for autonomous driving, including:
a learning-based dynamic modeling procedure, to enable the control-in-
the-loop simulation with highly accurate vehicle dynamics for parameter
tuning; a learning-based open-loop mapping procedure, to solve the
feedforward control parameters tuning; and a Bayesian-optimization (BO)-
based closed-loop parameter tuning procedure, to automatically tune
feedback control (PID, LQR, MRAC, MPC, etc.) parameters in simulation
environment.
• This framework has been validated on different vehicles in US and China.
29. A Learning-Based Tune-Free Control Framework for
Large Scale Autonomous Driving System Deployment
Complete tune-free control framework
30. A Learning-Based Tune-Free Control Framework for
Large Scale Autonomous Driving System Deployment
Simulation-based autonomous driving system
31. A Learning-Based Tune-Free Control Framework for
Large Scale Autonomous Driving System Deployment
Closed-loop control architecture
Linear-Quadratic- Regulator
= LQR
Model-Predictive-Control
= MPC
Model-Reference-Adaptive-Control
= MRAC
Proportional-Integral-Difference
= PID
32. A Learning-Based Tune-Free Control Framework for
Large Scale Autonomous Driving System Deployment
Simulation-based autotuner flowchart
A cloud-based simulation environment that runs autonomous driving full stack (from perception to
control) and, on this basis, distributes tens of suggested optimal parameters in thousands of
scenarios in parallel to generate running records/history under these parameters. 2.) A performance
profiling module that grades the simulation records generated from previous part. 3.) A Bayesian-
based optimizer that efficiently suggests new trials of parameters based on previous grading scores.
This finishes one iteration and sends back to simulation environment again for next iteration.
34. A Learning-Based Tune-Free Control Framework for
Large Scale Autonomous Driving System Deployment
Closed-loop control architecture
35. A Learning-Based Tune-Free Control Framework for
Large Scale Autonomous Driving System Deployment
Gaussian Process Regression = GPR
Tree-structured Parzen Estimator = TPE
Experimental vehicles in Apollo Platform