Enabling Courteous Vehicle Interactions through Game-Based and Dynamics-Aware Intent Inference

This article concerns the open question of creating control policies of autonomous vehicles (AVs) that lead to courteous motion. The study is built on a two-agent interaction between two agents (M and H), where each agent plans its motion by optimizing a trade-off of goal fulfillment, safety, and courtesy losses. The paper has three contributions: First, the 'double-blindness' issue in intent inference, i.e., inferring H's intent requires knowledge about H's inference of M's intent, is addressed. An empathetic intent inference algorithm is proposed, where H's intent, along with its inference of M's intent, are jointly inferred. Second, vehicle dynamics is explicitly incorporated into intent inference to acknowledge its influence on decision making in driving by the drivers' knowledge about dynamical properties of surrounding vehicles. Lastly, a courtesy loss that leverages intent inference is introduced. This loss measures the expected additional loss to H caused by M's motion from a baseline where M behaves rationally and in favor of H. Simulation studies are conducted to demonstrate that (1) joint inference and knowledge about vehicle dynamics are important to intent decoding and motion planning, and (2) the proposed courtesy definition leads to more rational motions than those from an existing definition.