Research map / Probabilistic and scientific modeling / Dynamical systems
Inverse Dynamics Pretraining Learns Good Representations for Multitask Imitation
David Brandfonbrener, Ofir Nachum, Joan Bruna
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 |
|---|---|---|---|
| Pretrain with inverse dynamics instead of other objectives | Inverse dynamics (ID) · Six imitation datasets, finetuning size sweep | success rate | Improved |
| Scale pretraining data with ID | Inverse dynamics (ID) · Pretraining size sweep | success rate | Improved |
| Include finetuning contexts in pretraining data (in-distribution eval) (ablation) | ID and BC · Door, Kitchen, Metaworld ML45, R3M | success rate | Improved |
| Include finetuning contexts in pretraining data (ablation) | Other pretraining algorithms · Four discrete-context datasets | success rate | No clear effect |
| Use ID instead of BC when context is fully latent | ID · Pointmass, Pick+place, Kitchen | success rate | Improved |
| Use ID instead of BC when context is observable | ID · Door, ML45, R3M | success rate | No clear effect |
| Predict forward dynamics instead of inverse dynamics | Forward dynamics representations · Action prediction | train loss | Hurt |
| Use ImageNet-pretrained features | ImageNet features · Action prediction | validation loss | Hurt |
| Add multistep dynamics to the model (ablation) | ID and implicit forward dynamics · Point mass, pick and place, door | success rate | No clear effect |
Closest papers by meaning
- Robust Visual Imitation Learning with Inverse Dynamics Representations (AAAI 2024 · similarity 0.89)
- DynaMo: In-Domain Dynamics Pretraining for Visuo-Motor Control (NeurIPS 2024 · similarity 0.89)
- Latent Diffusion Planning for Imitation Learning (ICML 2025 · similarity 0.88)
- What's the Move? Hybrid Imitation Learning via Salient Points (ICLR 2025 · similarity 0.88)
- Imitation from Observations with Trajectory-Level Generative Embeddings (NeurIPS 2026 · similarity 0.88)
- LMAct: A Benchmark for In-Context Imitation Learning with Long Multimodal Demonstrations (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.
