New York University Abu Dhabi
Robots that learn their skills and keep their guarantees.
REACH Lab works on the methods robots need to acquire skills and carry them out reliably and safely. We combine control theory, differential geometry and machine learning. What a robot learns from people and from its own interaction then respects the structure of the space it moves in. It also keeps the guarantees it needs in contact with the world and with the people in it.
Work with us
We welcome enquiries from people who want to work on robot learning and control. There is no application form: write to us.
Prospective PhD students
If your interests sit in one of our research fields, tell us which one and what you would like to find out. PhD admission is through the university's graduate programs; write to us before you apply.
Postdoctoral researchers
If you hold, or are completing, a doctorate in robotics, control or machine learning, tell us what you would bring and what you would build here.
Undergraduate researchers
If you study at NYU Abu Dhabi and want research experience in robot learning and control, tell us what draws you to it.
Research fields
Trajectory optimization and the identification of robot dynamic parameters run underneath most of them.
Geometric and Riemannian robot learning
Learning and control are formulated directly on manifolds and Lie groups, not on flattened approximations of them. This keeps what a robot learns consistent with the structure of the space it moves in.
Safe and certifiable learning-based control
We establish stability and safety guarantees for robot behaviour. These guarantees still hold when the components producing the behaviour were learned from data rather than designed.
Learning from demonstration and physical interaction
Robots acquire skills from human demonstration or from their own exploration. These skills then transfer across objects, goals and embodiments.
Compliant and contact-rich manipulation
We regulate force and impedance in tasks defined by contact. The aim is to keep the robot stable and safe while it touches an object or a person.
Soft and continuum robotics
We model, simulate and control bodies whose compliance is structural rather than commanded. Their very many degrees of freedom put them out of reach of classical rigid-body methods.
Decision-making under uncertainty
We use free-energy and active-inference formulations to treat exploration and robustness as one problem. This lets a robot act well without an exact model of the world.
Neural interfaces and assistive robotics
We decode human intent from muscle and brain activity. We then close the loop through prosthetic and wearable devices that respond to it.
News
people Narayana Reddy Marthala joined the lab as a PhD student.
people Dean Pretorius joined the lab as a PhD student.
paper “MV-UMI: A Scalable Multi-View Interface for Cross-Embodiment Learning” was accepted in IEEE Access. DOI
paper “K-VARK: Kernelized Variance-Aware Residual Kalman Filter for Sensorless Force Estimation in Collaborative Robots” was accepted in IEEE Transactions on Automation Science and Engineering. DOI
award “Escaping the Trap: Benchmarking Swarm Gradient-Following in Geometrically Constrained Environments” received a Best Video Award at ANTS 2026 in Darmstadt, and was runner-up for the Best Poster Award on the conference’s first day. DOI
Recent publications
Safe and Stable Neural Network Dynamical Systems for Robot Motion Planning
IEEE Robotics and Automation Letters (RA-L), vol. 11, no. 3, pp. 3111–3118
K-VARK: Kernelized Variance-Aware Residual Kalman Filter for Sensorless Force Estimation in Collaborative Robots
IEEE Transactions on Automation Science and Engineering (T-ASE)
MV-UMI: A Scalable Multi-View Interface for Cross-Embodiment Learning
IEEE Access
Escaping the Trap: Benchmarking Swarm Gradient-Following in Geometrically Constrained Environments
Lecture Notes in Computer Science (Springer) — ANTS 2026, 15th International Conference on Swarm Intelligence, Darmstadt








