Decision-Making, Dialogue & AI

Interesting behaviors and dialogues can be achieved with classical programming, with models like Behavior Trees or State Machines. But there is a lot to gain by bringing in AI techniques:

• Language models — including small, embeddable ones — and semantic extraction.
• Task planning (e.g. PDDL) to produce more flexible behaviors and dialogues.
• Situational pattern recognition and reinforcement learning.

My PhD in Human-Robot Interaction (ISIR, Sorbonne Université) was about teaching robots new behaviors using spoken language, and I applied it to production: a PDDL-based decision system running embedded on the Pepper robot, and natural interaction driven by language models on the Mirokai robots.

I favor explainable solutions that keep you in control, and the reduction of data collection.

What I can do for you

• Bring task planning, chatbots or machine learning into your product.
• Design dialogue and interaction systems grounded in HRI research.
• Keep the AI explainable, controllable and frugal with data.

Schedule a call to discuss your project.