Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Self-Predictive Representations for Combinatorial Generalization in Behavioral Cloning
Daniel Lawson, Adriana Hugessen, Charlotte Cloutier, Glen Berseth, Khimya Khetarpal
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 BYOL-gamma auxiliary representation loss to GCBC | GCBC · OGBench stitch average-all | success rate (%): 26 -> 35 | Improved |
| add minimal 1-step BYOL auxiliary loss | GCBC · OGBench stitch average-all | success rate (%): 26 -> 26 | No clear effect |
| add BYOL-gamma on visual environments | GCBC · OGBench average-visual | success rate (%): 34 -> 37 | Improved |
| add TD-SR auxiliary loss | GCBC · OGBench average-visual | success rate (%): 34 -> 31 | Hurt |
| use predictive horizon gamma>0 instead of gamma=0 (ablation) | BYOL-gamma · OGBench stitch average-all | success rate (%): 24 -> 33 | Improved |
| use predictive horizon gamma>0 instead of gamma=0 (ablation) | BYOL-gamma · humanoidmaze-medium-stitch | success rate (%): 18 -> 54 | Improved |
| use cross-entropy loss instead of normalized squared l2 (ablation) | BYOL-gamma · OGBench stitch average-all | success rate (%): 31 -> 33 | Improved |
| add action conditioning to the forward predictor (ablation) | BYOL-gamma · OGBench stitch average-all | success rate (%): 33 -> 33 | No clear effect |
| add backward predictor (ablation) | BYOL-gamma · OGBench stitch average-all | success rate (%): 33 -> 33 | No clear effect |
| use BYOL-gamma instead of TD-SR on large-state environments | TD-SR · humanoidmaze-medium-stitch | success rate (%): 42 -> 51 | Improved |
| use TD-SR instead of BYOL-gamma on small antmaze | BYOL-gamma · antmaze-large-stitch | success rate (%): 23 -> 19 | Hurt |
Closest papers by meaning
- Rethinking Latent Redundancy in Behavior Cloning: An Information Bottleneck Approach for Robot Manipulation (ICML 2025 · similarity 0.88)
- Self-Supervised Keyframe Discovery for Horizon-Invariant Behavior Cloning (NeurIPS 2026 · similarity 0.86)
- Is Behavior Cloning All You Need? Understanding Horizon in Imitation Learning (NeurIPS 2024 · similarity 0.86)
- Foundation Policies with Hilbert Representations (ICML 2024 · similarity 0.86)
- Improving Generative Behavior Cloning via Self-Guidance and Adaptive Chunking (NeurIPS 2025 · similarity 0.86)
- Diffusion Model-Augmented Behavioral Cloning (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.
