Research map / Reinforcement learning / Imitation learning
Learning Soft Constraints From Constrained Expert Demonstrations
Ashish Gaurav, Kasra Rezaee, Guiliang Liu, Pascal Poupart
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 |
|---|---|---|---|
| Use ICL instead of the strongest baseline, ICRL | Constraint recovery · Gridworld (A) | CMSE: 0.11 ± 0.02 -> 0.08 ± 0.01 | Improved |
| Use ICL instead of the strongest baseline, ICRL | Constraint recovery · Gridworld (B) | CMSE: 0.21 ± 0.04 -> 0.04 ± 0.01 | Improved |
| Use ICL instead of the strongest baseline, GAIL-Constraint | Constraint recovery · CartPole (MR) | CMSE: 0.12 ± 0.03 -> 0.02 ± 0.00 | Improved |
| Use ICL instead of the strongest baseline, GAIL-Constraint | Constraint recovery · CartPole (Mid) | CMSE: 0.25 ± 0.02 -> 0.08 ± 0.05 | Improved |
| Use ICL instead of the strongest baseline | Constraint recovery · Ant-Constrained and HalfCheetah-Constrained | CMSE: 0.17 ± 0.04 -> 0.07 ± 0.00; 0.20 ± 0.03 -> 0.05 ± 0.00 | Improved |
| Use ICL instead of the strongest baseline | Policy accrual matching · HalfCheetah-Constrained | NAD: 7.50 ± 4.97 -> 10.16 ± 7.49 | Hurt |
| Use ICL instead of the strongest baseline | Policy accrual matching · Ant-Constrained | NAD: 8.02 ± 2.84 -> 6.84 ± 1.29 | Improved |
| Add policy mixture and reweighting to constraint adjustment (ablation) | ICL · Gridworld (A) | CMSE | Improved |
| Add reweighting to the policy mixture (ablation) | ICL with a policy mixture · Gridworld (B) and CartPole (Mid) | convergence speed | Improved |
| Add reweighting to the policy mixture (ablation) | ICL with a policy mixture · CartPole (MR) | convergence speed | No clear effect |
| Increase transition stochasticity from deterministic to p_slip=0.1 (ablation) | ICL · Gridworld (A) | CMSE: 0.02 ± 0.00 | Improved |
| Increase transition stochasticity to p_slip=0.5 (ablation) | ICL · Gridworld (A) | CMSE: 0.08 ± 0.00 | Hurt |
Closest papers by meaning
- Recent Advancements in Inverse Reinforcement Learning (AAAI 2024 · similarity 0.91)
- Simplifying Constraint Inference with Inverse Reinforcement Learning (NeurIPS 2024 · similarity 0.90)
- Provably Efficient Exploration in Inverse Constrained Reinforcement Learning (ICML 2025 · similarity 0.90)
- Is Inverse Reinforcement Learning Harder than Standard Reinforcement Learning? A Theoretical Perspective (ICML 2024 · similarity 0.90)
- Offline Inverse RL: New Solution Concepts and Provably Efficient Algorithms (ICML 2024 · similarity 0.89)
- Sub-optimal Experts mitigate Ambiguity in Inverse Reinforcement Learning (NeurIPS 2024 · similarity 0.89)
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.
