A talk from the Develop Track at AWE USA 2017 - the largest conference for AR+VR in Santa Clara, California May 31- June 2, 2017.
Jeff Powers (Occipital): The Commoditization of Positional Tracking
As virtual reality and mixed reality systems reach broader adoption, more companies require the need for positional tracking to precisely match user movement with rendered content. Yet what initially began as a key competitive advantage has quickly evolved into a complex yet commonplace commodity. Lighthouse has been open-sourced. Inside-out positional tracking can be done with consumer-grade sensors. And future mobile devices will likely come with much of the sensor package required to enable 6-DoF positional tracking. While some aspects of positional tracking are being commoditized, others are still difficult open challenges. We'll explore the differences between visual odometry and full SLAM, sparse point clouds and dense models, and costs associated with different kinds of systems.
http://AugmentedWorldExpo.com
2. Confidential - Do Not Distribute
40+ strong
7 former startup founders
3 original Kinect hackers
TEAM PAST WORK
KEY BACKERS
THE TEAM
San Francisco, CA Boulder, CO
Lynx
ACQUISITIONS
3. We build hardware for 3D vision.
But today we’re talking about positional tracking software.
5. The standard of 1969: Oldsmobile
+ Spacious
+ Luxurious
+ Stylish (in 1969)
+ Powerful
- Expensive
- Inefficient
1969 Oldsmobile Cutlass Supreme Holiday Coupe
6. Along comes the Corolla
• 42% cheaper
+ 50% more efficient
• Fewer parts: manual, no
power steering, no AC
(Less to break)
• 54% smaller
1969 Toyota Corolla
7. So, how did it end?
2017 Toyota Corolla
• World’s best selling car by 1974
• World’s best selling nameplate by 1997
• 40 million sold worldwide
Oldsmobile
• Declining sales led to GM
shutting down Oldsmobile in
December 2000
12. The future of positional
tracking will rely on SLAM
and world-facing cameras.
13. ✅ No base stations or setup
✅ Multi-user anywhere via shared SLAM maps
✅ Automatic “chaperone” using the map
Why SLAM-based tracking?
14. ❌ No margin for error in algorithms
❌ Expensive: multi-camera, lasers, etc.
❌ Compute-heavy
So why isn’t SLAM-based tracking pervasive?
15. So, we set out to try to address
some of those challenges.
16. We started with our RGBD sensor. Depth sensing with multiple
optics and a dedicated ASIC. Too expensive for widest adoption.
Cameras (3): $15 - $60
IMU: $2-$5
Laser Chip: $5-$30
17. Stereo fisheye cameras + IMU:
Nature’s design, but still too costly to land in every device.
$10-20
$1-5
18. What about a single camera?
This could pave the way for 6-DoF
tracking in almost every AR/VR device.
$1-5
$5-10
21. Accuracy: very good, constantly improving.
8.5 mm
RMSE
0.48 °
Includes global
scale correction.
Room-scale test vs. Lighthouse
Lighthouse
Occipital
26. But what will it cost?
✅ Obstacle Awareness✅ 6-DoF ✅ Scale Accurate
27. $10 $7
Up To 10K
Licenses
10K-100K
Licenses
100K+
Licenses
28. FREE $10 $7
Up To 10K
Licenses*
10K-100K
Licenses
100K+
Licenses
*3% for devices with an MSRP > $700
Our basic positional tracking
software is free for up to 10K licenses.
Request access now at occipital.com/tracking
Currently on
35. Commoditization of Positional Tracking
• Cheap positional tracking for every device? Yes.
• Full visual AI & scene understanding? Not for a while.
• Today people are scrambling to get 6-DoF, but a year from
now it’ll be trying to level-up environmental awareness.