2016 D-STOP Symposium ("Smart Cities") session by CTR's Christian Claudel. Get symposium details: http://ctr.utexas.edu/research/d-stop/education/annual-symposium/
Networks of wearables and augmented reality for vulnerable user protection
1. Networks of wearables and
augmented reality for
vulnerable user protection
Christian Claudel
Assistant Professor
CAEE Department
University of Texas, Austin
D-STOP Annual Symposium 2016
2. Motivations
• Very high number of road accidents
- In the world, 1.3 million deaths/year
- 20-50 million injuries/year
• 90% of all crashes are caused by human error
• An increasing number of crashes occurs in cities,
with a high concentration of heterogeneous users
(cars, bicycles, pedestrians)
• State of the art: policy, planning, but no active safety
3. Proposed system
• Autonomous vehicles will help address the problem,
but will take time to ramp up
• Even with autonomous vehicles, we need
coordinated response by humans (ex: pedestrians,
bicyclists): human in the loop control
• Wearables (smart glasses, smart watches) can be a
solution to the problem:
- Low cost (distributed among users)
- Will penetrate market faster than autonomous vehicles
- Augmented reality capabilities for actuation
4. Vision
• Users wearing smart glasses (and other wearables)
• Positioning through higher high resolution GPS (RTK-GPS), or UAVs
equipped with cameras
5. Vision
• Learning-based framework to detect user intent and predict future
actions based on video and wearable inertial/positioning data
• Collision detection using predicted reachable sets, and resolution using
real-time path visualization (through augmented reality)
7. Challenges
• Sensing (using phone, glass, watch)
- IMUs for head and wrist tracking
- Cameras for scene detection/user intent detection
- RTK-GPSs for position estimation
• Networking
- Need for a low latency communication channel (Bluetooth, DSRC,
5G?)
• User path forecast and collision avoidance
- Motion tracking with machine learning for path forecast
- Collaborative collision avoidance with uncertain (human) actuation
– requires human factor experts
• Cybersecurity
- GPS spoofing
- Need for sensor fusion
8. Future work
• Currently: Google Glasses interfaced with RTK-GPS,
head motion tracking
• Future work:
- use ML to forecast user paths, develop collision detection
and avoidance algorithms
- use R7 glasses (with better field of view)
- Use of drones to replace RTK GPSs in the near future
(and provide data for path forecasts algorithms)
• Study human response
• Investigate for other uses: traffic intersection
management with mixed autonomous/human vehicles