Research map / Detection and segmentation / Tracking
RSPT: Reconstruct Surroundings and Predict Trajectory for Generalizable Active Object Tracking
Fangwei Zhong, Xiao Bi, Yudi Zhang, Wei Zhang, Yizhou Wang
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 structure reconstruction to target visibility input (ablation) | RSPT · Six unseen environments (Mean) | Success Rate: 0.45 -> 0.79 | Improved |
| Add trajectory prediction to target visibility input (ablation) | RSPT · Six unseen environments (Mean) | Accumulated Reward: 159 -> 277 | Improved |
| Combine structure reconstruction and trajectory prediction (ablation) | RSPT · Six unseen environments (Mean) | Accumulated Reward: 277 -> 302 | Improved |
| Replace MDN-RNN with a Kalman Filter (ablation) | RSPT · Six unseen environments (Mean) | Accumulated Reward: 302 -> 277 | Hurt |
| Use target-only binary masks instead of object masks (ablation) | Grounded-state tracker · Six unseen environments (Mean) | Accumulated Reward: 210 -> 343 | Improved |
| Replace estimated poses with grounded poses (ablation) | RSPT · Six unseen environments (Mean) | Accumulated Reward: 302 -> 366 | Improved |
| Add noise to depth input (ablation) | RSPT · Six unseen environments (Mean) | Accumulated Reward: 302 -> 281 | Hurt |
| Add noise to pose input (ablation) | RSPT · Six unseen environments (Mean) | Accumulated Reward: 302 -> 291 | Hurt |
| Add noise to both depth and pose inputs (ablation) | RSPT · Six unseen environments (Mean) | Accumulated Reward: 302 -> 270 | Hurt |
| Use RSPT rather than VT + Planner | Real-world tracker · Moving Forward | Success Rate: 0.90 -> 0.97 | Improved |
| Use RSPT rather than VT + Planner | Real-world tracker · Moving Backward | Success Rate: 0.83 -> 0.91 | Improved |
| Use RSPT rather than VT + Planner | Real-world tracker · Visual Occluding | Success Rate: 0.68 -> 0.95 | Improved |
Later papers that compared with this method
1 later accepted paper reports RSPT as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Instance-level Visual Active Tracking with Occlusion-Aware Planning (CVPR 2026 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Pose-Assisted Multi-Camera Collaboration for Active Object Tracking (AAAI 2020 · similarity 0.86)
- AD-VAT: An Asymmetric Dueling mechanism for learning Visual Active Tracking (ICLR 2019 · similarity 0.86)
- End-to-End Active Object Tracking and Its Real-World Deployment via Reinforcement Learning (TPAMI 2019 · similarity 0.85)
- Towards Generalizable Multi-Object Tracking (CVPR 2024 · similarity 0.85)
- ETAP: Event-based Tracking of Any Point (CVPR 2025 · similarity 0.84)
- S2-Track: A Simple yet Strong Approach for End-to-End 3D Multi-Object Tracking (ICML 2025 · similarity 0.83)
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.
