Research map / Reinforcement learning / Visual and representation RL
Affordances From Human Videos as a Versatile Representation for Robotics
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 VRB affordance-guided offline data collection instead of prior affordance models | offline imitation learning · Cabinet, Knife, Veg, Shelf, Pot, Door, Lid, Drawer | success rate: VRB k-NN: 0.6, 0.3, 0.6, 0.8, 0.4, 1.0, 0.4, 1.0; VRB behavior cloning: 0.6, 0.1, 0.3, 0.3, 0.8, 0.9, 0.2, 0.9 | Mixed |
| Use VRB rather than R3M visual features | frozen-encoder behavior cloning · Franka Kitchen: microwave, slide-door, door-open | success rate: microwave: 0.10 -> 0.16; slide-door: 0.70 -> 0.84; door-open: 0.11 -> 0.13 | Helped |
| Use VRB affordances rather than baseline affordance models | offline data collection in simulation · Franka Kitchen: Light, Microwave, Kettle | success rate: Light: 0.60 -> 0.75; Microwave: 0.45 -> 0.60; Kettle: 0.45 -> 0.55 | Helped |
| Use VRB rather than Hotspots affordances | grasping held-out objects · VR Controller, Chain, Stapler | grasping success rate: VR Controller: 0.13 -> 0.27; Chain: 0.20 -> 0.33; Stapler: 0.20 -> 0.53 | Helped |
| Use affordance-guided exploration rather than random or HAP-based exploration | reward-free exploration · Cabinet, Knife, Stovepot, Shelf | coincidental success: from 3× to 10× | Helped |
| Use VRB sampling rather than HAP or random sampling | goal-conditioned learning · Door, Veggies, Lid, Dishwasher, Drawer, Garbage Can | success rate | Helped |
| Use a VRB-derived discrete action space rather than a baseline-derived action space | DQN goal-conditioned control · Cabinet and Veggies | success | Helped |
Closest papers by meaning
- VidBot: Learning Generalizable 3D Actions from In-the-Wild 2D Human Videos for Zero-Shot Robotic Manipulation (CVPR 2025 · similarity 0.90)
- AffordDP: Generalizable Diffusion Policy with Transferable Affordance (CVPR 2025 · similarity 0.90)
- Affordance-R1: Reinforcement Learning for Generalizable Affordance Reasoning in Multimodal Large Language Models (AAAI 2026 · similarity 0.89)
- Robots Pre-train Robots: Manipulation-Centric Robotic Representation from Large-Scale Robot Datasets (ICLR 2025 · similarity 0.89)
- Mitigating the Human-Robot Domain Discrepancy in Visual Pre-training for Robotic Manipulation (CVPR 2025 · similarity 0.89)
- IAAO: Interactive Affordance Learning for Articulated Objects in 3D Environments (CVPR 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.