Research map / Reinforcement learning / Imitation learning
Object-Aware Regularization for Addressing Causal Confusion in Imitation Learning
Jongjin Park, Younggyo Seo, Chang Liu, Li Zhao, Tao Qin, Jinwoo Shin, Tie-Yan Liu
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 |
|---|---|---|---|
| Add object-aware masking that drops feature-map units sharing VQ-VAE discrete codes | Behavioral cloning · Confounded Atari (27 environments) | Mean HNS: 91.7% -> 105.6% | Improved |
| Add object-aware masking that drops feature-map units sharing VQ-VAE discrete codes | Behavioral cloning · Original Atari (27 environments) | Mean HNS: 99.0% -> 114.9% | Improved |
| Add object-aware masking that drops feature-map units sharing VQ-VAE discrete codes | Behavioral cloning · CARLA daytime tasks | Success rate: BC: 75.0 -> 87.0 (Straight); 43.0 -> 70.0 (One turn); 16.9 -> 35.7 (Navigation); 18.0 -> 30.0 (Navigation w/ dynamic obs | Improved |
| Use fixed VQ-VAE representations instead of an action-trained encoder initialized from VQ-VAE (ablation) | OREO behavioral cloning · Confounded Atari (8 selected environments) | Mean HNS: 100.1% -> 41.3% | Hurt |
| Replace object-aware masking with Dropout on fixed VQ-VAE representations (ablation) | OREO behavioral cloning · Confounded Atari (8 selected environments) | Mean HNS: 100.1% -> 45.7% | Hurt |
| Replace object-aware masking with DropBlock on fixed VQ-VAE representations (ablation) | OREO behavioral cloning · Confounded Atari (8 selected environments) | Mean HNS: 100.1% -> 53.0% | Hurt |
| Increase the per-code drop probability (ablation) | OREO · Confounded Atari (8 environments) | Mean HNS | Improved |
| Use a codebook that is too small or too large (ablation) | OREO · Confounded Atari (8 environments) | Mean HNS | Hurt |
| Train with object-aware masking rather than plain behavioral cloning (ablation) | OREO · Confounded Atari across varying demonstration sizes | Mean HNS | Improved |
| Train the VQ-VAE from a very small demonstration set (N=5) (ablation) | OREO · Confounded Atari (8 environments) | Mean HNS | Mixed |
| Apply object-aware masking consistently across four stacked observations | Behavioral cloning · Confounded Atari (8 environments, four stacked observations) | Mean HNS: 62.3% -> 87.4% | Improved |
| Add OREO regularization to DRIL's behavioral-cloning policies | DRIL · Confounded CrazyClimber and Pong | Sample efficiency | Improved |
Closest papers by meaning
- Videos are Sample-Efficient Supervisions: Behavior Cloning from Videos via Latent Representations (NeurIPS 2025 · similarity 0.88)
- Is Behavior Cloning All You Need? Understanding Horizon in Imitation Learning (NeurIPS 2024 · similarity 0.88)
- Beyond-Expert Performance with Limited Demonstrations: Efficient Imitation Learning with Double Exploration (ICLR 2025 · similarity 0.87)
- Efficient Imitation under Misspecification (ICLR 2025 · similarity 0.87)
- OLLIE: Imitation Learning from Offline Pretraining to Online Finetuning (ICML 2024 · similarity 0.87)
- Rethinking Latent Redundancy in Behavior Cloning: An Information Bottleneck Approach for Robot Manipulation (ICML 2025 · 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.
