Research map / Reinforcement learning / Visual and representation RL
Hardware Conditioned Policies for Multi-Robot Transfer Learning
Tao Chen, Adithyavairavan Murali, Abhinav Gupta
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 |
|---|---|---|---|
| Condition the policy on explicit robot kinematics | DDPG+HER policy · Unseen robot types, reacher and peg insertion | success rate: 92.50±1.96 vs 0.20±0.40; 92.20±2.75 vs 0.00±0.00 | Improved |
| Condition the policy on a learned hardware embedding | PPO on hopper · Hopper with varied kinematics and dynamics | return | Improved |
| Use learned hardware embedding instead of explicit kinematics-only encoding | HCP on hopper · Hopper with varied kinematics and dynamics | return | Improved |
| Pretrain policy weights on other hoppers, then fine-tune on new hoppers | HCP-I · New hopper agents | learning speed | Improved |
| Pretrain the HCP-E policy on other robot types, then fine-tune on the new type | HCP-E · Peg insertion on new robot types H and I | learning speed | Improved |
| Add ground-truth dynamics parameters to the policy input (ablation) | DDPG+HER · 7-DOF reacher and fixed-goal peg insertion | learning speed | Hurt |
| Add a random vector to the policy input (ablation) | DDPG+HER · 7-DOF reacher and fixed-goal peg insertion | learning speed | Hurt |
| Train on damping values [0.01, 2), then test on [30, 40) | DDPG+HER · Peg insertion | success rate: 100 % -> 85 % | Hurt |
| Train on damping values [0.01, 2), then test on [40, 50) | DDPG+HER · Peg insertion | success rate: 100 % -> 47 % | Hurt |
Closest papers by meaning
- Policy Transfer with Strategy Optimization (ICLR 2019 · similarity 0.87)
- Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes (AAAI 2025 · similarity 0.86)
- Scaling Proprioceptive-Visual Learning with Heterogeneous Pre-trained Transformers (NeurIPS 2024 · similarity 0.85)
- Deep Policy Gradient Methods Without Batch Updates, Target Networks, or Replay Buffers (NeurIPS 2024 · similarity 0.85)
- Zero-Shot Robotic Manipulation with Pre-Trained Image-Editing Diffusion Models (ICLR 2024 · similarity 0.85)
- Foundation Policies with Hilbert Representations (ICML 2024 · similarity 0.85)
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.
