Multi-modal sketch-based behavior tree program automatic generation technology

What is it about?

Behavior trees are hierarchical, structured, and modular domain-specific languages (DSL). Constructing desired behavior tree programs is challenging. We propose an automatic behavior tree synthesis framework, BtBot, to help non-expert users generate BT programs that meet given specifications. Users input a natural language description of a task and positive and negative examples as the specifications. BtBot employs LLMs to generate a sketch. BtBot completes the sketch using enumeration search technology. If a sketch is infeasible, it repairs the sketch using a multi-step repair technique. Experiments demonstrate that BtBot outperforms BtGenBot, a technique for fine-tuning LLM to generate behavior trees, and other baseline techniques. Finally, user studies demonstrate that BtBot effectively helps both expert and non-expert users build behavior tree programs.

Featured Image

Photo by Possessed Photography on Unsplash

Photo by Possessed Photography on Unsplash

Why is it important?

This is a great application of program synthesis technology in the field of behavior trees. The key innovations of this paper are the abstraction of behavior tree programs and the repair of behavior tree sketches guided by counterexamples.

Perspectives