Research map / Label-efficient and robust learning / Self-supervised and contrastive
Learn TAROT with MENTOR: A Meta-Learned Self-Supervised Approach for Trajectory Prediction
Mozhgan Pourkeshavarz, Changhe Chen, Amir Rasouli
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add TAROT-prediction self-supervised HSL tasks balanced by MENTOR (ablation) | HGT trajectory predictor · Argoverse turns | DAO: 68.45 -> 79.21 | Improved |
| add a single TAROT proxy task (length 6) (ablation) | HGT trajectory predictor · Argoverse turns | DAO: 68.45 -> 76.93 | Improved |
| increase TAROT length from 6 to 8 (ablation) | single-TAROT HSL task · Argoverse turns | DAO: 76.93 -> 73.65 | Hurt |
| combine multiple TAROT lengths via MENTOR instead of a single best TAROT (ablation) | HSL tasks · Argoverse turns | DAO: 76.93 -> 79.21 | Improved |
| run four SSL tasks through MENTOR instead of adding them individually (ablation) | lane-graph baseline · Argoverse validation | DAO: 69.97 -> 77.91 | Improved |
| add SSL task B (distance to intersection) alone (ablation) | lane-graph baseline · Argoverse validation | RF: 3.00 -> 3.87 | Improved |
| add SSL task A (lane masking) alone (ablation) | lane-graph baseline · Argoverse validation | RF: 3.00 -> 2.98 | No clear effect |
| use MENTOR instead of HiVT on challenging turn scenarios | trajectory prediction · Argoverse turns | DAO: 66.84 -> 79.21 | Improved |
| use MENTOR instead of HiVT on the full Argoverse test set | trajectory prediction · Argoverse test | minFDE: 1.16 -> 1.21 | Hurt |
| remove loss-based weighting of proxy tasks (naive multi-task combination) (ablation) | multi-task HSL training · Argoverse | primary task performance | Hurt |
Closest papers by meaning
- Self-Supervised Cross-City Trajectory Representation Learning Based on Meta-Learning (AAAI 2026 · similarity 0.87)
- SeqGrowGraph: Learning Lane Topology as a Chain of Graph Expansions (ICCV 2025 · similarity 0.86)
- LaneSegNet: Map Learning with Lane Segment Perception for Autonomous Driving (ICLR 2024 · similarity 0.85)
- RoadSceneBench: A Lightweight Benchmark for Mid-Level Road Scene Understanding (CVPR 2026 · similarity 0.84)
- MoST: Multi-Modality Scene Tokenization for Motion Prediction (CVPR 2024 · similarity 0.84)
- LaKD: Length-agnostic Knowledge Distillation for Trajectory Prediction with Any Length Observations (NeurIPS 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
