Research map / Reinforcement learning / Policy gradient and actor-critic
Guided Cost Learning: Deep Inverse Optimal Control via Policy Optimization
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 GCL instead of relative entropy IRL | dish placement NN cost · PR2 dish rack | success rate: 0\% -> 100\% | Helped |
| use GCL instead of relative entropy IRL | pouring NN cost · PR2 pouring | success rate: 10\% -> 84.7\% | Helped |
| affine cost instead of neural network cost (ablation) | GCL · PR2 pouring | success rate: 84.7\% -> 0\% | Hurt |
| reoptimize a new policy with the learned cost instead of using GCL policy (ablation) | GCL pouring · PR2 pouring new positions | success rate: 84.7\% -> 34\% | Hurt |
| reoptimize with learned cost on new target positions (ablation) | GCL dish · PR2 dish rack | success rate: 100\% -> 100\% | No clear effect |
| estimate demo importance weights empirically instead of ground truth (ablation) | GCL · point mass consistency | KL divergence: 230.66 -> 272.71 | Hurt |
| drop demo and sample importance weights (ablation) | GCL · point mass consistency | variance of recovered distribution | Hurt |
| remove maximum entropy term from policy optimization (ablation) | GCL · point mass consistency | variance of recovered distribution: variance 0 | Hurt |
| remove local constant rate regularization (ablation) | GCL · simulated reaching and peg insertion | final distance | Hurt |
| remove monotonic regularization (ablation) | GCL · simulated reaching and peg insertion | final distance | Hurt |
| add more suboptimal samples for prior methods | relative entropy IRL and others · simulated tasks | performance | No clear effect |
| learn cost on image pixels | GCL · PR2 pouring | success | Hurt |
Closest papers by meaning
- Optimal Exploration for Model-Based RL in Nonlinear Systems (NeurIPS 2023 · similarity 0.87)
- Efficient Exploration of Reward Functions in Inverse Reinforcement Learning via Bayesian Optimization (NeurIPS 2020 · similarity 0.87)
- Hybrid Inverse Reinforcement Learning (ICML 2024 · similarity 0.87)
- Meta-Inverse Reinforcement Learning with Probabilistic Context Variables (NeurIPS 2019 · similarity 0.87)
- Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation (ICLR 2025 · similarity 0.86)
- ASID: Active Exploration for System Identification in Robotic Manipulation (ICLR 2024 · 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.