Research map / Reinforcement learning / Imitation learning
Where Do You Think You're Going?: Inferring Beliefs about Dynamics from Behavior
Sid Reddy, Anca Dragan, Sergey Levine
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 learned internal dynamics for internal-to-real dynamics transfer assistance | Human Lunar Lander control · Lunar Lander user study | Success rate: f(1,11)=109.58, p<0.0001 | Improved |
| Use learned internal dynamics for internal-to-real dynamics transfer assistance | Human Lunar Lander control · Lunar Lander user study | Crash rate: f(1,11)=126.33, p<0.0001 | Improved |
| Use learned internal dynamics to assist control | Lunar Lander user study · Lunar Lander user study | Average speed: 30% decrease | Improved |
| Train on more than one known-reward task (ablation) | Inverse soft Q-learning · Grid World Navigation | Internal dynamics prediction accuracy | Improved |
| Add the action intent prior restricting predictions to states reachable under real dynamics (ablation) | Inverse soft Q-learning · Grid World Navigation with globally scrambled actions | Internal dynamics prediction accuracy | Improved |
| Add the action intent prior restricting predictions to states reachable under real dynamics (ablation) | Inverse soft Q-learning · Grid World Navigation with locally scrambled actions | Internal dynamics prediction accuracy | Mixed |
| Use constraint sampling and deterministic internal dynamics approximations for continuous states (ablation) | Inverse soft Q-learning · 2D Continuous Navigation | Internal dynamics parameter error | No clear effect |
| Use learned internal dynamics rather than assume internal dynamics equal real dynamics | MaxCausalEnt IRL · Grid World reward learning from misguided demonstrations | True reward collected by learned-reward policy | Improved |
| Use inverse soft Q-learning instead of simultaneous reward-and-dynamics estimation | SERD baseline · Grid World reward learning from misguided demonstrations | Reward learning performance | Improved |
| Model user behavior using learned internal dynamics instead of real dynamics | MaxCausalEnt IRL · Grid World reward learning from misguided demonstrations | Reward learning performance | Improved |
Closest papers by meaning
- Quantifying the Sensitivity of Inverse Reinforcement Learning to Misspecification (ICLR 2024 · similarity 0.90)
- When Your AIs Deceive You: Challenges of Partial Observability in Reinforcement Learning from Human Feedback (NeurIPS 2024 · similarity 0.87)
- Estimating cognitive biases with attention-aware inverse planning (NeurIPS 2025 · similarity 0.87)
- Inverse decision-making using neural amortized Bayesian actors (ICLR 2025 · similarity 0.87)
- Partial Identifiability in Inverse Reinforcement Learning for Agents with Non-Exponential Discounting (AAAI 2025 · similarity 0.87)
- Behavior-agnostic Task Inference for Robust Offline In-context Reinforcement Learning (ICML 2025 · similarity 0.87)
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.
