Research map / Reinforcement learning / Visual and representation RL
Dual-Agent Reinforcement Learning for Adaptive and Cost-Aware Visual-Inertial Odometry
Feiyang Pan, Shenghe Zheng, Chunyan Yin, Guangbin Dou
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 |
|---|---|---|---|
| replace heuristic fixed-weight fusion with RL-trained Fusion Agent (ablation) | Dual-agent VIO · EuRoC MH04 | ATE (m): 0.143 -> 0.112 | Improved |
| replace EKF fusion with RL-trained Fusion Agent (ablation) | Dual-agent VIO · EuRoC MH04 | ATE (m): 0.127 -> 0.112 | Improved |
| add the learned Fusion Agent to DROID-VO frontend (ablation) | DROID-VO · EuRoC MH04 | ATE (m): 0.399 -> 0.237 | Improved |
| estimate both gyro and accelerometer bias with IMU Bias Estimator (ablation) | Dual-agent VIO · EuRoC test clips and final framework | ATE | Improved |
| use fixed bias estimation instead of random noise model (ablation) | IMU Bias Estimator · EuRoC final framework | ATE | Improved |
| use IMU-only RL scheduling instead of heuristic gating at aggressive skip (75-87.5%) (ablation) | Select Agent · EuRoC ATE vs skip ratio | ATE | Improved |
| add visual inputs to the scheduler (RL-gating KF) instead of IMU-only (ablation) | Select Agent · EuRoC, 50% skip target | throughput and ATE: ATE increase < 3% for IMU-only, with higher FPS | No clear effect |
| compare scheduling policies at 50% skip ratio (ablation) | Select Agent · EuRoC | ATE | No clear effect |
| add Select Agent to skip VO frames | Dual-agent VIO vs DPVO · EuRoC | throughput (FPS): 22 -> 39 | Improved |
| replace VIBA-heavy backend with RL agents (vs ORB-SLAM3) | Dual-agent VIO · EuRoC | CPU BA/VIBA time (ms): 121.09 -> 12.77 | Improved |
| replace VIBA-heavy backend with RL agents (vs ORB-SLAM3) | Dual-agent VIO · EuRoC (CPU-based comparison) | ATE (m): 0.041 -> 0.092 | Hurt |
| add 5-10% blurred and noisy frames | Dual-agent VIO vs DPVO · EuRoC MH04 | ATE (m): 0.112 -> 0.153 (ours), 0.137 -> 0.192 (DPVO) | Mixed |
Closest papers by meaning
- Adaptive VIO: Deep Visual-Inertial Odometry with Online Continual Learning (CVPR 2024 · similarity 0.90)
- Tartan IMU: A Light Foundation Model for Inertial Positioning in Robotics (CVPR 2025 · similarity 0.86)
- Seeing through Uncertainty: Robust Task-Oriented Optimization in Visual Navigation (NeurIPS 2025 · similarity 0.85)
- EqNIO: Subequivariant Neural Inertial Odometry (ICLR 2025 · similarity 0.85)
- LEAP-VO: Long-term Effective Any Point Tracking for Visual Odometry (CVPR 2024 · similarity 0.85)
- HASARD: A Benchmark for Vision-Based Safe Reinforcement Learning in Embodied Agents (ICLR 2025 · 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.
