# CASD: Chunk-Aligned Semantic Distillation for Multi-StageRobot Manipulation

arXiv: https://arxiv.org/abs/2609.08638 · HTML: https://mlsanity.com/paper/2609.08638/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 16 of its 36 references that are in the map.
- First posted on arXiv 8 Sep 2026.

## Builds on (most cited first, 16 of 16)

- [Do As I Can, Not As I Say: Grounding Language in Robotic Affordances](https://mlsanity.com/paper/2204.01691/) (2022)
- [π0.5: a Vision-Language-Action Model with Open-World Generalization](https://mlsanity.com/paper/2504.16054/) (2025)
- [Code as Policies: Language Model Programs for Embodied Control](https://mlsanity.com/paper/2209.07753/) (2022)
- [RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization for Robust Bimanual Robotic Manipulation](https://mlsanity.com/paper/2506.18088/) (2025)
- [Causal World Modeling for Robot Control](https://mlsanity.com/paper/2601.21998/) (2026)
- [Motus: A Unified Latent Action World Model](https://mlsanity.com/paper/2512.13030/) (2025)
- [RT-H: Action Hierarchies Using Language](https://mlsanity.com/paper/2403.01823/) (2024)
- [Compositional Foundation Models for Hierarchical Planning](https://mlsanity.com/paper/2309.08587/) (2023)
- [Qwen-RobotManip Technical Report: Alignment Unlocks Scale for Robotic Manipulation Foundation Models](https://mlsanity.com/paper/2606.17846/) (2026)
- [SARM: Stage-Aware Reward Modeling for Long Horizon Robot Manipulation](https://mlsanity.com/paper/2509.25358/) (2025)
- [Universal Visual Decomposer: Long-Horizon Manipulation Made Easy](https://mlsanity.com/paper/2310.08581/) (2023)
- [ProgressVLA: Progress-Guided Diffusion Policy for Vision-Language Robotic Manipulation](https://mlsanity.com/paper/2603.27670/) (2026)
- [ProcVLM: Learning Procedure-Grounded Progress Rewards for Robotic Manipulation](https://mlsanity.com/paper/2605.08774/) (2026)
- [Training Vision-Language-Action Models with Dense Embodied Chain-of-Thought Supervision](https://mlsanity.com/paper/2606.30552/) (2026)
- [Revisiting Embodied Chain-of-Thought for Generalizable Robot Manipulation](https://mlsanity.com/paper/2606.03784/) (2026)
- [Privileged Foresight Distillation: Zero-Cost Future Correction for World Action Models](https://mlsanity.com/paper/2604.25859/) (2026)

## Abstract (arXiv)

An action chunk can span several stages of a manipulation task, yet a label for its first step describes only the current stage. We introduce Chunk-Aligned Semantic Distillation (CASD), which derives semantic targets for entire action chunks. An offline vision--language model segments demonstrations into described stages. Their occupancy within each action chunk determines a weighted semantic target, including transitions between stages. A CASD generator learns to predict this target from the current observation, robot state, and task instruction. We then freeze the generator and train a policy conditioned on its predictions. The semantic branch runs once per policy query, without online VLM calls or reasoning-trace decoding. Teacher matching on annotated LIBERO training episodes is above chance for both single-stage and boundary-crossing chunks. We evaluate three Fast-WAM variants and a DreamZero integration across four benchmarks, including distribution shifts on LIBERO-Plus. Compared with published references, IDM+CASD reaches 98.9% versus 98.0% average success on LIBERO, while Uncond falls below its reference. Joint+CASD reaches 93.0% versus 90.6% on RoboTwin 2.0, and DreamZero+CASD reaches a 47.9% four-category MolmoSpaces manipulation average versus 40.7%. Performance varies across backbone integrations.