Research map / Reinforcement learning / Imitation learning
Language-Conditioned Imitation Learning for Robot Manipulation Tasks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 language-conditioned attention and closed-loop motor-primitive control instead of the Full RNN baseline | Robot manipulation imitation learning · Simulation test environments, closed loop | sequence success rate: 0.00 -> 0.84 | Helped |
| use language-conditioned attention and closed-loop motor-primitive control instead of PayAttention! | Robot manipulation imitation learning · Simulation test environments, closed loop | sequence success rate: 0.00 -> 0.84 | Helped |
| add attention to trajectory-only training (ablation) | Language-conditioned imitation model · Ablation test environments | sequence success rate: 0.28 -> 0.51 | Helped |
| add the W, phase, and phase-progression losses without attention (ablation) | Language-conditioned imitation model · Ablation test environments | sequence success rate: 0.28 -> 0.00 | Hurt |
| replace attention with a feed-forward network over all image features (ablation) | Language-conditioned imitation model · Ablation test environments | sequence success rate: 0.84 -> 0.00 | Hurt |
| replace motor primitives with a recurrent network predicting the next joint configuration (ablation) | Language-conditioned imitation model · Ablation test environments | sequence success rate: 0.84 -> 0.00 | Hurt |
| replace motor primitives with feed-forward step prediction (ablation) | Language-conditioned imitation model · Ablation test environments | sequence success rate: 0.84 -> 0.79 | Hurt |
| increase training data from 2,500 to 30,000 samples (ablation) | Language-conditioned imitation model · Ablation test environments | sequence success rate: 0.10 -> 0.80 | Helped |
| increase training data from 30,000 to 40,000 samples (ablation) | Language-conditioned imitation model · Ablation test environments | sequence success rate: 0.80 -> 0.84 | Helped |
| use free-form commands from previously unseen human users | Our model · New-user instructions | sequence success rate: 0.84 -> 0.64 | Hurt |
| test under changed illumination | Our model · Illumination perturbation | sequence success rate: 0.84 -> 0.62 | Hurt |
Closest papers by meaning
- Learning with Language-Guided State Abstractions (ICLR 2024 · similarity 0.91)
- Mastering Robot Manipulation with Multimodal Prompts through Pretraining and Multi-task Fine-tuning (ICML 2024 · similarity 0.89)
- Grounding Language Plans in Demonstrations Through Counterfactual Perturbations (ICLR 2024 · similarity 0.89)
- Inner Speech as Behavior Guides: Steerable Imitation of Diverse Behaviors for Human-AI coordination (NeurIPS 2025 · similarity 0.89)
- Policy Improvement using Language Feedback Models (NeurIPS 2024 · similarity 0.89)
- Instant Policy: In-Context Imitation Learning via Graph Diffusion (ICLR 2025 · 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.