Research map / Reinforcement learning / Imitation learning
NavOL: Navigation Policy with Online Imitation Learning
Xiaofei Wei, Chun Gu, Li Zhang
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 online imitation learning with expert-in-the-loop rollouts on top of NavDP | NavDP · NavOL benchmark (8 held-out scenes) | mSR: 42.2 -> 69.0 | Improved |
| add online imitation learning on top of NavDP | NavDP · NavDP benchmark | mSR: 77.8 -> 80.4 | Improved |
| add online imitation learning on top of NavDP | NavDP · NavDP benchmark Intern-Home | SR: 60.0 -> 58.8 | Hurt |
| execute expert actions with probability 0.2 during rollout (ablation) | NavOL · NavOL benchmark | SR: 43.9 -> 69.0 | Improved |
| initialize from pretrained NavDP weights (ablation) | NavOL · NavOL benchmark | SR: 48.3 -> 69.0 | Improved |
| train on more scenes (1 -> 10 -> 50) (ablation) | NavOL · NavOL benchmark | SR: 57.8 -> 64.8 -> 69.0 | Improved |
| vary rollout length per iteration (8, 32 vs default 128 steps) (ablation) | NavOL · NavOL benchmark | SR: 67.9 (8 steps), 68.5 (32 steps) vs 69.0 | No clear effect |
| add path post-processing (local line search for clearance) in the expert planner (ablation) | NavOL · NavOL benchmark | SR: 64.2 -> 69.0 | Improved |
| add camera height and pitch randomization (ablation) | NavOL · NavOL benchmark | SR: 66.0 -> 69.0 | Improved |
| add critic for test-time trajectory selection (ablation) | NavOL · NavOL benchmark | SR: 67.1 -> 69.0 | Improved |
| deploy on real Unitree Go2 zero-shot instead of NavDP | NavOL · Real-world Office/Gym/Corridor | success rate: Office 4/10 -> 8/10; Gym 2/10 -> 7/10; Corridor 2/10 -> 10/10 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DD-PPO — DD-PPO: Learning Near-Perfect PointGoal Navigators from 2.5 Billion Frames (ICLR 2020) · better on 2 of 2 shared results
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
- Is Behavior Cloning All You Need? Understanding Horizon in Imitation Learning (NeurIPS 2024 · similarity 0.89)
- Prior Does Matter: Visual Navigation via Denoising Diffusion Bridge Models (CVPR 2025 · similarity 0.89)
- RLIF: Interactive Imitation Learning as Reinforcement Learning (ICLR 2024 · similarity 0.88)
- Policy Decorator: Model-Agnostic Online Refinement for Large Policy Model (ICLR 2025 · similarity 0.88)
- What's the Move? Hybrid Imitation Learning via Salient Points (ICLR 2025 · similarity 0.88)
- Reward-free World Models for Online Imitation Learning (ICML 2025 · similarity 0.88)
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.
