Research map / Continual learning / Online and replay-based
Lifelong Imitation Learning with Multimodal Latent Replay and Incremental Adjustment
Fanqi Yu, Matteo Tiezzi, Tommaso Apicella, Cigdem Beyan, Vittorio Murino
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 IFA loss on top of latent replay (ablation) | MLR · LIBERO-OBJECT | AUC: 77.6 -> 79.4 | Improved |
| add IFA loss on top of latent replay (ablation) | MLR · LIBERO-GOAL | AUC: 74.6 -> 77.2 | Improved |
| add IFA loss on top of latent replay (ablation) | MLR · LIBERO-50 | AUC: 54.7 -> 56.1 | Improved |
| add IFA loss on top of latent replay (ablation) | MLR · LIBERO-OBJECT | NBT: 12.3 -> 11.4 | No clear effect |
| select IFA pairs by language + agent-view similarity instead of language + action (ablation) | MLR + IFA · LIBERO-OBJECT | AUC: 75.1 -> 79.4 | Improved |
| use top 50% of task pairs instead of 66.6% (ablation) | MLR + IFA · LIBERO-OBJECT | AUC: 75.8 -> 79.4 | Improved |
| use language embedding as reference instead of mean global latent (ablation) | MLR + IFA · LIBERO-GOAL | AUC: 70.5 -> 77.2 | Improved |
| store features with probability 0.50 instead of 0.10 (ablation) | MLR + IFA · LIBERO-GOAL | AUC: 66.2 -> 77.2 | Improved |
| store features with probability 0.50 instead of 0.20 (ablation) | MLR + IFA · LIBERO-GOAL | AUC: 77.4 -> 77.2 | No clear effect |
| use angular distance instead of cosine distance in IFA loss (ablation) | MLR + IFA · LIBERO-OBJECT and LIBERO-GOAL | FWT, NBT, AUC | Improved |
| replay latent features instead of raw trajectories, frozen CLIP | MLR · LIBERO-GOAL | AUC: 60.5 -> 74.6 | Improved |
Closest papers by meaning
- Incremental Learning of Retrievable Skills For Efficient Continual Task Adaptation (NeurIPS 2024 · similarity 0.89)
- CCIL: Continuity-Based Data Augmentation for Corrective Imitation Learning (ICLR 2024 · similarity 0.88)
- Latent Diffusion Planning for Imitation Learning (ICML 2025 · similarity 0.88)
- InTAct: Interval-based Task Activation Consolidation for Continual Learning (NeurIPS 2026 · similarity 0.87)
- Think Small, Act Big: Primitive Prompt Learning for Lifelong Robot Manipulation (CVPR 2025 · 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.
