Offline Training: We highlight our behavior-guided navigation policy for autonomous driving. We use a behavior-rich simulator that can generate aggressive or conservative driving styles. In Step 1, we use the CMetric behavior classification algorithm to compute a set of parameters that characterize aggressive behaviors such as over-speeding, overtaking, and sudden lane changing. In Step 2, we use these parameters to train a behavior-based action class navigation policy for action prediction and local navigation.
Runtime: We use our behavior-guided trained policy and the final simulation parameters computed using offline training. During an episode at runtime, we use the trained policy to predict the next action of the ego-vehicle given the current state of the traffic environment, which is represented in the form of a feature matrix. The predicted action (in this case, “turn left”) is converted into the final local trajectory using the internal controls of the simulator, modified by the parameters that take into account the behavior of traffic agents.
| Paper | Code | Supplementary Material |
|---|---|---|
| B-GAP | GitHub Code | Video |
We present an algorithm for behaviorally-guided action prediction and local navigation for autonomous driving in dense traffic scenarios. Our approach classifies the driver behavior of other vehicles or road-agents (aggressive or conservative) and considers that information for decision-making and safe driving. We present a behavior-driven simulator that can generate trajectories corresponding to different levels of aggressive behaviors, and we use our simulator to train a reinforcement learning policy using a multilayer perceptron neural network. We use our reinforcement learning-based navigation scheme to compute safe trajectories for the ego-vehicle accounting for aggressive driver maneuvers such as overtaking, over-speeding, weaving, and sudden lane changes. We have integrated our algorithm with the OpenAI gym-based ``Highway-Env'' simulator and demonstrate the benefits of our navigation algorithm in terms of reducing collisions by 3.25−26.90% and handling scenarios with 2.5× higher traffic density.
Please cite our work if you found it useful,
@article{mavrogiannis2021bgap,
title={B-GAP: Behavior-Guided Action Prediction and Navigation for Autonomous Driving},
author={Mavrogiannis, Angelos and Chandra, Rohan and Manocha, Dinesh},
journal={arXiv preprint arXiv:2011.03748},
year={2021}
}