Conditional Prediction by Simulation for Automated Driving
Fabian Konstantinidis, Moritz Sackmann, Ulrich Hofmann, Christoph Stiller
TL;DR
This work tackles the problem of cooperative planning in automated driving by developing a conditional prediction framework that couples trajectory planning with interactive motion prediction. It leverages Adversarial Inverse Reinforcement Learning to learn a realistic driver policy and embeds this policy in a closed-loop, graph-based multi-agent simulator to generate predictions conditioned on a candidate ego trajectory. The approach enables bidirectional interactions, allowing the ego plan to influence predicted behaviors and vice versa, while supporting dynamic adaptation during prediction rollouts. Evaluations on the INTERACTION dataset show robust, scene-consistent predictions and demonstrate conditional prediction capabilities at intersections, highlighting potential for improving cooperative planning and reducing the freezing robot problem. Future work includes integrating a more capable planning module to fully exploit the conditional prediction framework for safe, cooperative driving.
Abstract
Modular automated driving systems commonly handle prediction and planning as sequential, separate tasks, thereby prohibiting cooperative maneuvers. To enable cooperative planning, this work introduces a prediction model that models the conditional dependencies between trajectories. For this, predictions are generated by a microscopic traffic simulation, with the individual traffic participants being controlled by a realistic behavior model trained via Adversarial Inverse Reinforcement Learning. By assuming various candidate trajectories for the automated vehicle, we generate predictions conditioned on each of them. Furthermore, our approach allows the candidate trajectories to adapt dynamically during the prediction rollout. Several example scenarios are available at https://conditionalpredictionbysimulation.github.io/.