Counterfactual Policy Evaluation for Decision-Making in Autonomous Driving
This work addresses safety and generalization issues in autonomous driving decision-making, though it is incremental as it builds on existing learning-based approaches with a novel evaluation framework.
The paper tackles the problem of learned policies in autonomous driving failing to generalize to novel situations by introducing a counterfactual policy evaluation method that assesses policies in hypothetical scenarios where other agents behave differently, resulting in a significant decrease in collision rates while maintaining high success rates.
Learning-based approaches, such as reinforcement and imitation learning are gaining popularity in decision-making for autonomous driving. However, learned policies often fail to generalize and cannot handle novel situations well. Asking and answering questions in the form of "Would a policy perform well if the other agents had behaved differently?" can shed light on whether a policy has seen similar situations during training and generalizes well. In this work, a counterfactual policy evaluation is introduced that makes use of counterfactual worlds - worlds in which the behaviors of others are non-actual. If a policy can handle all counterfactual worlds well, it either has seen similar situations during training or it generalizes well and is deemed to be fit enough to be executed in the actual world. Additionally, by performing the counterfactual policy evaluation, causal relations and the influence of changing vehicle's behaviors on the surrounding vehicles becomes evident. To validate the proposed method, we learn a policy using reinforcement learning for a lane merging scenario. In the application-phase, the policy is only executed after the counterfactual policy evaluation has been performed and if the policy is found to be safe enough. We show that the proposed approach significantly decreases the collision-rate whilst maintaining a high success-rate.