Safety and Efficiency in Autonomous Vehicles through Online Learning
Zachary Sunberg
What's the most important objective for an autonomous system?
Safety
What's the enemy of safety?
Uncertainty
What happens when we deploy autonomous systems? (Even when we have safety in mind)
Shoettle and Sivak, "A Preliminary Analysis of Real-World Crashes Involving Self-Driving Vehicles" UMTRI-2015-34
737 Max
Uber Crash
Two Objectives for Autonomy
EFFICIENCY
SAFETY
Minimize resource use
(especially time)
Minimize the risk of harm to oneself and others
Safety often opposes Efficiency
Pareto Optimization
Safety
Better Performance
Model \(M_2\), Algorithm \(A_2\)
Model \(M_1\), Algorithm \(A_1\)
Efficiency
$$\underset{\pi}{\mathop{\text{maximize}}} \, V^\pi = V^\pi_\text{E} + \lambda V^\pi_\text{S}$$
Safety
Weight
Efficiency
Tweet by Nitin Gupta
29 April 2018
https://twitter.com/nitguptaa/status/990683818825736192
Overconfident MDP
Conservative MDP
Multiple lane change simulation results
[Sunberg, 2017]
Types of Uncertainty
OUTCOME
MODEL
STATE
All drivers normal
No learning (MDP)
Omniscient
Simulation results
[Sunberg, 2017]
POMCP
POMCP-DPW
POMCPOW
(Silver, 2010)
(Sunberg, 2017)
(Sunberg, 2018)
All drivers normal
No learning (MDP)
Omniscient
POMCPOW (Ours)
Simulation results
[Sunberg, 2017]
Next
- New Algorithms
- New Sources of Uncertainty
- POMDPs in the Real World
New Algorithms
New Sources of Uncertainty
POMDPs in the real world
Thank You!
AUT Talks
By Zachary Sunberg
