Research

This project addresses the problem of grasping and pushing objects in unstructured, uncertain and cluttered environments. Such environment is found in daily human activities and we will inevitably need to face such challenge in order to achieve autonomous robotic manipulation in human environments. This research thus aims at leveraging physics-based predictive models within optimisation frameworks to overcome these challenges. Our most recent results will be presented in Humanoids-18 (arxiv.org/abs/1805.03005) and WAFR-18 (arxiv.org/abs/1807.09049)
Check out our demo video
We adopt the view of leveraging interaction with obstacles in the environment to achieve the semantic goal of a manipulation task. The focus of this project is on learning deep policies for pushing objects in a cluttered environment to a target region with minimal disturbance to the final environment state. We envision a closed-loop behaviour that is robust to uncertainty and can control the robot in real time. This work encompasses the development of novel techniques in imitation learning, short and long horizon planning, and deep reinforcement learning. Our most recent result will be presented in Humanoids-18 (arxiv.org/abs/1803.08100)
Check out our demo video
This project focuses on the manipulation planning for human-robot forceful collaboration. We are aiming at developing a general-purpose manipulation planning framework, where robots are capable of assisting humans in performing forceful tasks (e.g., drilling, cutting) in a safe, comfortable and natural manner. This project is at the intersection of robotic manipulation and grasping, multi-robot system, motion planning and human-robot collaboration. Check our most recent paper in IEEE/RSJ IROS 2018 (http://eprints.whiterose.ac.uk/133934/)
Check out our demo video
Grasp and motion planning for human-robot comfortable forceful interaction
This project addresses the problem of grasp planning and objects positioning during forceful human-robot physical collaboration. There are infinite ways a robot can hold an interacting object in task-space during physical HRI, yet we are interested in finding a global position which is comfortable for the human to interact and operate---without compromising grasp stability. We are now exploring offline strategies based on human kinematics and muscle activation, but in the future, we will also consider reactiveness and real-time motion and adaptation to human posture/behaviour. Our most recent result will be presented in Humanoids-18 (arxiv.org/abs/1807.11323)