TrustMHE monitors AI prediction errors online and blends AI-based planning costs with a constant-velocity fallback, significantly reducing crashes in closed-loop simulation.
MGTR: Multi-Granular Transformer for Motion Prediction with LiDAR
1 Pith paper cite this work. Polarity classification is still indexing.
abstract
Motion prediction has been an essential component of autonomous driving systems since it handles highly uncertain and complex scenarios involving moving agents of different types. In this paper, we propose a Multi-Granular TRansformer (MGTR) framework, an encoder-decoder network that exploits context features in different granularities for different kinds of traffic agents. To further enhance MGTR's capabilities, we leverage LiDAR point cloud data by incorporating LiDAR semantic features from an off-the-shelf LiDAR feature extractor. We evaluate MGTR on Waymo Open Dataset motion prediction benchmark and show that the proposed method achieved state-of-the-art performance, ranking 1st on its leaderboard (https://waymo.com/open/challenges/2023/motion-prediction/).
fields
cs.RO 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Enhancing System Self-Awareness and Trust of AI: A Case Study in Trajectory Prediction and Planning
TrustMHE monitors AI prediction errors online and blends AI-based planning costs with a constant-velocity fallback, significantly reducing crashes in closed-loop simulation.