Research

My research is dedicated to enabling decision autonomy in robotic systems. This involves creating robots capable of perceiving, planning, and executing tasks independently, without human supervision.

Research Interests

Within this broader vision, my focus is on the control architectures for legged robots, addressing unique challenges such as:

• Complex and Multiple-DOF Systems: Legged robots exhibit multiple degrees of freedom (DOF) and hybrid discrete states due to their leg movements (e.g., picking up and putting down legs).
• Dynamic Stability Challenges: These systems are rarely statically stable due to dynamic motion, resulting in non quasi-static movement where the centre of mass is not always above the support polygon. Balancing becomes a key control problem.
• Uncertain and Dynamic Environments: Operating in highly dynamic environments with uncertain conditions such as rough terrain and variable environmental contexts.

Current Research Areas

1. Deep Reinforcement Learning: Developing biologically-inspired, modular and decentralised control systems to improve exploration and generalisation in high-dimensional action spaces.
2. Simulation to Real-World Transfer: Ensuring robust transfer of learned policies from simulations to real-world scenarios, addressing deviations and uncertainties not captured in simulated environments.
3. Unsupervised Deep Learning: Deriving low-dimensional state spaces from high-dimensional sensory inputs, integrating temporal and behavioural relevance to enhance reinforcement learning outcomes.