CoRL 2026 · Memory for Robot Foundation Models
Submissions closed September 30, 2026 · Notifications by October 24, 2026
We invited short papers on any aspect of memory for robot foundation models. Position papers, works-in-progress, benchmark and dataset contributions, and negative or ablation results were explicitly welcomed.
Recurrent and state-space memory · compressed or gated KV caches · external memory banks · retrieval-augmented policies · on-robot latency and compute budgets
Persistent 3D and 4D scene representations · semantic maps · object-centric and surfel memory · Gaussian splats · object permanence under occlusion · re-localization and revisit consistency
Recall of trajectories, failures and corrections · retrieval-based adaptation · consolidation into reusable skills · forgetting stale memories · catastrophic interference
Memory-augmented action-conditioned prediction · spatial and physical consistency over long rollouts · hallucination in revisited scenes · memory-conditioned planning · hierarchical task state
Genuinely partially observable, history-dependent tasks · temporal, spatial, object and procedural memory taxonomies · controls that rule out longer context, memorization and task shortcuts
Preferences, instructions and corrections across sessions and users · interpretable and editable memory · privacy, consent and the right to be forgotten