As autonomous vehicles populate the road, they need to interact with other autonomous and human drivers. Our goal is to create algorithms for autonomous vehicles that exhibit socially-compliant behavior, both in human-like driving and in interactions with human drivers.
Navigating Tricky Intersections
Intersections are risky for humans and autonomous vehicles alike. This work examined how to integrate uncertainty to estimate risk for an autonomous vehicle navigating tricky road junctions. It weighs several critical factors, including occlusions, noisy sensors, other vehicles, and attentiveness of other drivers. The vehicle then knows when it is safe to enter an intersection when the risk is low.
Risk Level Sets
In urban environments, these vehicles will need to navigate dense traffic situations, thus creating a need for planning in congestion. Our approach creates a dynamic control law that scales with the varying traffic density. We generate a cost function for the vehicle, which incorporates the density, occupancy, and risk level within the environment. By choosing vehicles actions only below a certain cost value, we demonstrate the vehicle remains safe while avoiding the common “freezing robot” problem. As we change the given risk level, we can tune the vehicle’s personality on the road, from conservative to aggressive.
We can also apply this to autonomous wheelchairs