BAAI presents CausGround, a causal planner that guides real robots without retraining

AIBAAI presents CausGround, which uses simulation-distilled causal knowledge to guide real dual-arm robots without retraining. In the system, a frozen MLLM interprets natural language into a goal state, and CausGround reasons backward through a causal graph to generate valid action sequences, such as Open(Microwave) ➔ Pick(Bread) ➔ Place(Bread, Plate).

BAAIOfficial

@BAAIBeijing

View on X

Part of a thread · earlier post

Can simulation-distilled causal knowledge directly guide real robots without retraining?

BAAI‘s answer from CausGround is YES.

In real-world dual-arm manipulation tasks, the frozen MLLM only interprets natural language into a goal state. CausGround reasons backward through the causal graph to automatically generate valid sequences—like Open(Microwave) ➔ Pick(Bread) ➔ Place(Bread, Plate)!

Explore our interactive demos, benchmark, and paper:
🌐 Project:https://causground.github.io
📄 arXiv： https://arxiv.org/abs/2610.29702

Source details

BAAI · x.comPublished Nov 5, 8:46 AM ET

Original title: Can simulation-distilled causal knowledge directly guide real robots without retraining?

BAAI presents CausGround, a causal planner that guides real robots without retraining

Topics