Uncertainty-aware Short-term Motion Prediction of Traffic Actors for Autonomous Driving
Nemanja Djuric, Vladan Radosavljevic, Henggang Cui, Thi Nguyen, Fang-Chieh Chou, Tsung-Han Lin, Nitin Singh, Jeff Schneider
TL;DR
The paper tackles short-term motion prediction for traffic actors around autonomous vehicles by rasterizing actor-centric scenes with high-definition map context and feeding them to CNNs to predict future positions and associated uncertainties. It introduces an uncertainty-aware loss based on a half-normal model to capture aleatoric uncertainty and optionally uses an LSTM decoder for improved temporal modeling. The approach integrates dynamic state and map information, achieves strong empirical performance across multiple CNN backbones, and demonstrates well-calibrated uncertainty via reliability analyses and case studies. Extensive real-world data and onboard testing validate its practical potential for safer, more reliable SDV planning.
Abstract
We address one of the crucial aspects necessary for safe and efficient operations of autonomous vehicles, namely predicting future state of traffic actors in the autonomous vehicle's surroundings. We introduce a deep learning-based approach that takes into account a current world state and produces raster images of each actor's vicinity. The rasters are then used as inputs to deep convolutional models to infer future movement of actors while also accounting for and capturing inherent uncertainty of the prediction task. Extensive experiments on real-world data strongly suggest benefits of the proposed approach. Moreover, following completion of the offline tests the system was successfully tested onboard self-driving vehicles.