Research map / Detection and segmentation / Tracking
End-to-End Active Object Tracking and Its Real-World Deployment via Reinforcement Learning
Wenhan Luo, Peng Sun, Fangwei Zhong, Wei Liu, Tong 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 environment augmentation (randomized initial positions and flips) to training (ablation) | Active tracker · ViZDoom Standard | AR: 840 -> 2547 | Improved |
| add environment augmentation (invisible background objects, varied start points) (ablation) | Active tracker · UE S2MP2 | AR: -453.4 -> 1203.6 | Improved |
| add the LSTM sequence encoder (ablation) | Active tracker · ViZDoom Standard | AR: 1522 -> 2547 | Improved |
| replace LSTM with GRU (ablation) | Active tracker · ViZDoom Standard | AR: 2547 -> 2022 | No clear effect |
| add three more convolutional layers (deeper network) (ablation) | Active tracker · ViZDoom Standard | AR: 2547 -> 2535 | No clear effect |
| use end-to-end active tracker instead of passive tracker plus PID-like camera control (MIL) | Active tracker · ViZDoom Standard | EL: 743.1 -> 2959 | Improved |
| use end-to-end active tracker instead of Meanshift with camera control | Active tracker · UE S2MP2 | EL: 3000.0 -> 2565.3 | Mixed |
| use discrete 9-action space instead of continuous actions in real robot (ablation) | Active tracker on TurtleBot · Real robot indoor | Success rate: 0.7 -> 1.0 | Improved |
| use discrete 9-action space instead of continuous actions in real robot (ablation) | Active tracker on TurtleBot · Real robot outdoor | Success rate: 0.7 -> 0.9 | Improved |
| change target appearance, background or distractor at test time | Active tracker · ViZDoom FloorCeiling | AR: 2547 -> 1504 | Hurt |
Closest papers by meaning
- End-to-end Active Object Tracking via Reinforcement Learning (ICML 2018 · similarity 0.98)
- Visual Tracking via Hierarchical Deep Reinforcement Learning (AAAI 2021 · similarity 0.88)
- Tracking as Online Decision-Making: Learning a Policy From Streaming Videos With Reinforcement Learning (ICCV 2017 · similarity 0.88)
- Collaborative Deep Reinforcement Learning for Multi-Object Tracking (ECCV 2018 · similarity 0.88)
- Learning Policies for Adaptive Tracking With Deep Feature Cascades (ICCV 2017 · similarity 0.88)
- Open-World Drone Active Tracking with Goal-Centered Rewards (NeurIPS 2025 · similarity 0.87)
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.
