SlotVLA: Towards Modeling of Object-Relation Representations in Robotic Manipulation
Tung Kieu, Kashu Yamazaki, Chase Rainwater, Anh Nguyen, Ngan Le
AI summary
Problem
Current vision-language-action models rely on dense visual embeddings that entangle objects with background noise, causing high computational costs and poor interpretability, while existing object-centric methods miss critical gripper-object relational cues.
Approach
SlotVLA compresses dense visual inputs into a small set of task-filtered object slots and relation tokens using slot attention, which are then decoded into actions by a language model.
Key results
- Introduction of LIBERO+, a benchmark with fine-grained object-relation annotations
- Drastic reduction in required visual tokens while maintaining competitive success rates
- Task-aware slot filtering successfully isolates relevant objects and gripper interactions
- Improved interpretability and generalization over dense and purely object-centric baselines
Why it matters
Enables scalable, interpretable, and computationally efficient visuomotor control for complex multitask robotic manipulation.
Abstract
Inspired by how humans reason over discrete objects and their relationships, we explore whether compact object-centric and object-relation representations can form a foundation for multitask robotic manipulation. Most exist- ing robotic multitask models rely on dense embeddings that entangle both object and background cues, raising concerns about both efficiency and interpretability. In contrast, we study object–relation-centric representations as a pathway to more structured, efficient, and explainable visuomotor control. Our contributions are two-fold. First, we introduce LIBERO+, a fine-grained benchmark dataset designed to enable and evaluate object-relation reasoning in robotic manipulation. Unlike prior datasets, LIBERO+ provides object-centric annotations that enrich demonstrations with box- and mask-level labels as well as instance-level temporal tracking, supporting compact and interpretable visuomotor representations. Second, we propose SlotVLA, a slot-attention–based framework that captures both objects and their relations for action decoding. It uses a slot-based visual tokenizer to maintain consistent temporal object representations, a relation-centric decoder to produce task-relevant embeddings, and an LLM-driven module that translates these embeddings into executable actions. Experi- ments on LIBERO+ demonstrate that object-centric slot and object-relation slot representations drastically reduce the num- ber of required visual tokens, while providing competitive generalization. Together, LIBERO+ and SlotVLA provide a compact, interpretable, and effective foundation for advancing object–relation-centric robotic manipulation. Our full project is publicly available at https://slot-vla.github.io.