Better Prediction for Reinforcement Learning in Robotics and Autonomous Driving
Reinforcement learning (RL) is the area of artificial intelligence research that has the goal of allowing autonomous agents to learn from trial-and-error interaction with a task environment. One of the key capabilities required of a reinforcement learning agent is the ability to make predictions about its environment. For example, the agent must be able to predict the effect of its actions on the world state or be able to predict the future behavior of other agents in the environment. In this talk I will describe recent work towards improving prediction for RL settings.
In the first part of the talk I’ll describe recent work on applying reinforcement learning to robots in simulated environments with the goal of transferring learned skills to the physical world. I’ll introduce a family of techniques that improve the simulator’s predictions of the effects of a robot’s actions. These techniques allow reinforcement learning to be entirely applied in simulated environments and the resulting skills transferred to physical robots. In the second part of the talk I’ll talk about predicting future trajectories for other agents with a focus on autonomous driving scenarios. I’ll describe how unobserved vehicles or pedestrians can confound the ability of an agent to infer the goals of other agents and then introduce an algorithm that improves accuracy of goal inference. More accurate goal inference improves prediction which leads to fewer collisions by a Monte Carlo tree search planner in autonomous driving scenarios.
Josiah Hanna is an assistant professor in the Computer Sciences Department at the University of Wisconsin — Madison. He received his Ph.D. in the Computer Science Department at the University of Texas at Austin. Prior to attending UT Austin, he completed his B.S. in computer science and mathematics at the University of Kentucky. Before joining UW–Madison, he was a post-doc at the University of Edinburgh and also spent time at FiveAI working on autonomous driving. Josiah is a recipient of the NSF Graduate Research Fellowship and the IBM Ph.D. Fellowship.
His research interests lie in artificial intelligence and machine learning, seeking to develop algorithms that allow autonomous agents to learn (efficiently) from experience. In particular, he studies reinforcement learning and methods to increase the data efficiency of reinforcement learning algorithms.