Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Gibson Env: Real-World Perception for Embodied Agents
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 real scanned spaces rather than synthetic scene databases | Gibson environment · Real-world depth estimation | Real-world transfer error: 2.89† -> 0.92§ | Helped |
| Add Goggles to map real images toward the rendered-image domain | Gibson view synthesis · Depth estimation, f(Is) to u(It) | Depth estimation error: 1.560 -> 0.915 | Helped |
| Add Goggles to map real images toward the rendered-image domain | Gibson view synthesis · Scene classification, f(Is) to u(It) | Class accuracy: 0.266 -> 0.291 | Helped |
| Jointly train forward rendering f and inverse mapping Goggles u | Gibson view synthesis · Appearance similarity between renderings and real images | SSIM: 0.480 -> 0.816 | Helped |
| Jointly train forward rendering f and inverse mapping Goggles u | Gibson view synthesis · Appearance similarity between renderings and real images | L1: 0.10 -> 0.051 | Helped |
| Give the agent visual depth input instead of non-visual sensors only | Husky local planning and obstacle avoidance · Local planning and obstacle avoidance | Reward | Helped |
| Give the agent visual input instead of non-visual state only | Distant visual navigation agent · Distant visual navigation | Task success | Helped |
| Give the agent visual input | Ant stair-climbing agent · Stair climb with shifted initial and target locations | Performance: 70% | Helped |
Closest papers by meaning
- Learning Interactive Real-World Simulators (ICLR 2024 · similarity 0.86)
- ACTIVE-o3 : Empowering MLLMs with Active Perception via Pure Reinforcement Learning (ICML 2026 · similarity 0.85)
- SimWorld Studio: Automatic Environment Generation with Evolving Coding Agent for Embodied Agent Learning (NeurIPS 2026 · similarity 0.85)
- Steve-Eye: Equipping LLM-based Embodied Agents with Visual Perception in Open Worlds (ICLR 2024 · similarity 0.85)
- Neural Motion Simulator Pushing the Limit of World Models in Reinforcement Learning (CVPR 2025 · similarity 0.85)
- Any2Policy: Learning Visuomotor Policy with Any-Modality (NeurIPS 2024 · 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.