Research map / Reinforcement learning / Visual and representation RL
Habitat 2.0: Training Home Assistants to Rearrange their Habitat
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 MonolithicRL rather than SPA | Pick · Unseen layouts | success rate: 86.3 vs 72.7 | Helped |
| Use MonolithicRL rather than SPA-Priv | Pick · Unseen layouts | success rate: 86.3 vs 80.0 | Helped |
| Generalize the MonolithicRL policy from training layouts to unseen layouts | MonolithicRL · Pick | success rate: 91.7 \rightarrow 86.3 | Hurt |
| Generalize the MonolithicRL policy to unseen objects | MonolithicRL · Pick | success rate: 91.7 \rightarrow 74.7 | Hurt |
| Generalize the MonolithicRL policy to unseen receptacles | MonolithicRL · Pick | success rate: 91.7 \rightarrow 52.7 | Hurt |
| Add rendering optimizations (ablation) | H2.0 · Simulation benchmark, all process and interaction conditions | simulation steps per second: See Table: Benchmarking H2.0 performance | Helped |
| Add physics optimizations (ablation) | H2.0 · Simulation benchmark, all process and interaction conditions | simulation steps per second: See Table: Benchmarking H2.0 performance | Helped |
| Add all simulator optimizations (ablation) | H2.0 · Simulation benchmark, all process and interaction conditions | simulation steps per second: See Table: Benchmarking H2.0 performance | Helped |
| Use proprioception without visual input (ablation) | MonolithicRL · Pick, unseen layouts | success rate: 78% success; force-threshold failures 2x more; wrong-object picks 3x more | Mixed |
| Use head and arm cameras instead of arm camera alone (ablation) | MonolithicRL · Pick, unseen layouts | success rate: 85% to 92% | Helped |
| Add a one-step observation delay (ablation) | H2.0 agent policies · Pick | performance | No clear effect |
| Use velocity control instead of end-effector control (ablation) | Pick skill · Pick | learning performance | No clear effect |
Closest papers by meaning
- Habitat 3.0: A Co-Habitat for Humans, Avatars, and Robots (ICLR 2024 · similarity 0.90)
- ManiSkill-HAB: A Benchmark for Low-Level Manipulation in Home Rearrangement Tasks (ICLR 2025 · similarity 0.89)
- SimWorld Studio: Automatic Environment Generation with Evolving Coding Agent for Embodied Agent Learning (NeurIPS 2026 · similarity 0.87)
- LabUtopia: High-Fidelity Simulation and Hierarchical Benchmark for Scientific Embodied Agents (NeurIPS 2025 · similarity 0.87)
- Embodied-R1.5: Evolving Physical Intelligence via Embodied Foundation Models (NeurIPS 2026 · similarity 0.86)
- VirtualEnv: A Platform for Embodied AI Research (AAAI 2026 · 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.