Kanghyun Ryu, Minjun Sung, Piyush Gupta, Jovin D'sa, Faizan M. Tariq, David Isele, Sangjae Bae
Honda Research Institute, UC Berkeley, UIUC
Motion planning for autonomous vehicles (AVs) in dense traffic is challenging, often leading to overly conservative behavior and unmet planning objectives. This challenge stems from the AV’s limited ability to anticipate and respond to the interactive behavior of surrounding agents. Traditional decoupled prediction and planning pipelines rely on non-interactive predictions that overlook the fact that agents often adapt their behavior in response to the AV’s actions. To address this, we propose Interaction-Aware Neural Network-Enhanced Model Predictive Path Integral (IANN-MPPI) control, which enables interactive trajectory planning by predicting how surrounding agents may react to each control sequence sampled by MPPI. To improve performance in structured lane environments, we introduce a spline-based prior for the MPPI sampling distribution, enabling efficient lane-changing behavior. We evaluate IANN-MPPI in a dense traffic merging scenario, demonstrating its ability to perform efficient merging maneuvers.
(1) When applying MPPI with a traditional prediction–planning stack, the fixed prediction does not account for how surrounding vehicles might adapt to the ego vehicle’s actions. As a result, lane change behaviors are often predicted as unsafe. (2) In contrast, our method generates different future trajectories conditioned on each MPPI sample. For instance, the neural network prediction may anticipate that surrounding vehicles will yield in response to the ego vehicle’s lane change. This conditional prediction enables the motion planner to make more efficient decisions.
In MPPI, control candidates are usually sampled from Gaussian distribution. However, this can be inefficient when optimal control follows multimodal distribution, such as autonomous driving where optimal control can be merging into neighboring lane or following current lane. In this work, we sample control candidates not only from typical Gaussian distribution, but also from around the spline trajectory merging into neighboring lanes. Integrated with Biased-MPPI [3], this spline curve aids efficient lane change behaviors.
Our IANN-MPPI predicts that neighboring vehicles may yield to the ego vehicles, resulting in consistent nudging trials and finally successfully merging into the target lane.
In addition, our IANN-MPPI also can make safe decision making even when surrounding vehicles are not yielding.
