THOMAS: Trajectory Heatmap Output with learned Multi-Agent Sampling
arXiv:2110.06607
Abstract
In this paper, we propose THOMAS, a joint multi-agent trajectory prediction framework allowing for an efficient and consistent prediction of multi-agent multi-modal trajectories. We present a unified model architecture for simultaneous agent future heatmap estimation, in which we leverage hierarchical and sparse image generation for fast and memory-efficient inference. We propose a learnable trajectory recombination model that takes as input a set of predicted trajectories for each agent and outputs its consistent reordered recombination. This recombination module is able to realign the initially independent modalities so that they do no collide and are coherent with each other. We report our results on the Interaction multi-agent prediction challenge and rank on the online test leaderboard.
References in corpus (19)
- INTERACTION Dataset: An INTERnational, Adversarial and Cooperative moTION Dataset in Interactive Driving Scenarios with Semantic Maps
- Stepwise Goal-Driven Networks for Trajectory Prediction
- TNT: Target-driveN Trajectory Prediction
- Argoverse: 3D Tracking and Forecasting with Rich Maps
- Multiple Futures Prediction
- Trajectory Forecasts in Unknown Environments Conditioned on Grid-Based Plans
- DESIRE: Distant Future Prediction in Dynamic Scenes with Interacting Agents
- Learning to Predict Vehicle Trajectories with Model-based Planning
- VectorNet: Encoding HD Maps and Agent Dynamics from Vectorized Representation
- ReCoG: A Deep Learning Framework with Heterogeneous Graph for Interaction-Aware Trajectory Prediction
- Map-Adaptive Goal-Based Trajectory Prediction
- What-If Motion Prediction for Autonomous Driving
- Rules of the Road: Predicting Driving Behavior with a Convolutional Model of Semantic Interactions
- DenseTNT: End-to-end Trajectory Prediction from Dense Goal Sets
- Probabilistic Multi-modal Trajectory Prediction with Lane Attention for Autonomous Vehicles
- LaneRCNN: Distributed Representations for Graph-Centric Motion Forecasting
- Asymmetrical Bi-RNN for pedestrian trajectory encoding
- LaPred: Lane-Aware Prediction of Multi-Modal Future Trajectories of Dynamic Agents
- Scene Transformer: A unified architecture for predicting multiple agent trajectories