Instance-Aware Predictive Navigation in Multi-Agent Environments
This addresses the problem of safe and explainable autonomous navigation in complex multi-agent settings, representing a novel method for a known bottleneck.
The paper tackles efficient end-to-end learning of driving policies in dynamic multi-agent environments by proposing an Instance-Aware Predictive Control (IPC) approach that forecasts agent interactions and scene structures, achieving a new state of the art in CARLA simulation without expert demonstration.
In this work, we aim to achieve efficient end-to-end learning of driving policies in dynamic multi-agent environments. Predicting and anticipating future events at the object level are critical for making informed driving decisions. We propose an Instance-Aware Predictive Control (IPC) approach, which forecasts interactions between agents as well as future scene structures. We adopt a novel multi-instance event prediction module to estimate the possible interaction among agents in the ego-centric view, conditioned on the selected action sequence of the ego-vehicle. To decide the action at each step, we seek the action sequence that can lead to safe future states based on the prediction module outputs by repeatedly sampling likely action sequences. We design a sequential action sampling strategy to better leverage predicted states on both scene-level and instance-level. Our method establishes a new state of the art in the challenging CARLA multi-agent driving simulation environments without expert demonstration, giving better explainability and sample efficiency.