Research map / Reinforcement learning / Visual and representation RL
VER: Scaling On-Policy RL Leads to the Emergence of Navigation in Embodied Rearrangement
Erik Wijmans, Irfan Essa, Dhruv Batra
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 |
|---|---|---|---|
| use variable experience rollouts instead of equal-length rollouts (ablation) | VER versus NoVER · Open Fridge, 8 GPUs | Mean throughput (SPS): 2216 ±60 -> 2861 ±21 | Improved |
| use VER instead of DD-PPO | distributed on-policy RL · PointNav, HM3D | Throughput (SPS): 3065 SPS to 5325 SPS | Improved |
| use VER instead of DD-PPO | distributed on-policy RL · ObjectNav, MP3D | Throughput (SPS): 1021 to 2019 | Improved |
| allow navigation actions during skill training (ablation) | TP-SRL skills · Tidy House | Success: 2% Success to 32% | Improved |
| train Pick and Place skills with navigation actions but omit a separate navigation skill | TP-SRL(NoNav) · NavPick | Success: 50% | Improved |
| train Pick and Place skills with navigation actions but omit a separate navigation skill | TP-SRL(NoNav) · NavPickNavPlace | Success: 20% | Improved |
| add navigation actions to skills (ablation) | TP-SRL skills · Prepare Groceries | Success: 0% to 5% | Improved |
| add navigation actions to skills (ablation) | TP-SRL skills · Set Table, first pick+place | Success: 0% to 7.5% | Improved |
| allow skills to navigate for recovery | TP-SRL(NoNav) · Prepare Groceries, first interaction | Success: -34% Success | Hurt |
| allow skills to navigate for recovery | TP-SRL(NoNav) · Set Table, first interaction | Success: -45% Success | Hurt |
| retain stale data from preempted rollouts for extra PPO epochs (ablation) | VER · Open Fridge, 8 GPUs | Sample efficiency | Improved |
| collect more experience from faster-simulating environments (ablation) | VER · ObjectNav, MP3D | Sample efficiency | 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 4 of 4 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
- HASARD: A Benchmark for Vision-Based Safe Reinforcement Learning in Embodied Agents (ICLR 2025 · similarity 0.87)
- Rapidly Adapting Policies to the Real-World via Simulation-Guided Fine-Tuning (ICLR 2025 · similarity 0.87)
- GenPO: Generative Diffusion Models Meet On-Policy Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Staggered Environment Resets Improve Massively Parallel On-Policy Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- METRA: Scalable Unsupervised RL with Metric-Aware Abstraction (ICLR 2024 · similarity 0.86)
- Video-Enhanced Offline Reinforcement Learning: A Model-Based Approach (ICML 2025 · similarity 0.86)
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.
