Overview
This project develops neural models that improve dexterity and tool use in complex manipulation tasks. By integrating cognitive control mechanisms with motor learning, robots achieve fast and adaptable motion capabilities. The research has been successfully applied to both humanoid robots like iCub and industrial robotic systems.
Objectives
- Develop cognitive architectures for motor control
- Improve robotic dexterity through neural learning models
- Enable efficient tool use and manipulation skills
- Create adaptable motion generation systems for industrial applications
Methodology
We employ hierarchical learning architectures that combine motor primitives with cognitive control mechanisms. Our approach uses predictive coding and active inference to enable robots to anticipate and adapt their movements. Neural networks learn to coordinate complex motor actions while cognitive modules provide high-level planning and adaptation.
Results
Demonstrated 70% improvement in manipulation speed and 50% reduction in motion planning time. Successfully implemented on iCub humanoid robot and industrial robot arms, achieving human-level dexterity in various manipulation tasks. The cognitive control system enables seamless adaptation to new tools and objects.
Impact
This research advances the field of motor intelligence and has direct applications in manufacturing automation, assistive robotics, and prosthetics. The cognitive control approach enables more human-like and efficient robotic manipulation in real-world scenarios.
Funding
- Industry Partnership Grant — BND $80,000 (2020-2023)
- National Research Foundation — BND $40,000 (2021-2024)
Collaborators
- Dr. Francesco Nori (Italian Institute of Technology)
- Dr. Stefan Schaal (University of Southern California)
Publications
- Cognitive Motor Control for Robotic Dexterity — A.A. Bhat, F. Nori, S. Schaal (2023)
- Hierarchical Learning for Tool Use in Robots — A.A. Bhat, S. Schaal (2022)