Research projects

Teleoperation of dexterous hands to then collect data to perform imitation learning is a hot topic recently. We approach this problem from the perspective of compliant hands. We are currently studying how could we simplify and make teleoperation more robust and precise to different operators.

Cooking is a particularly interesting manipulation task with a large variety in task requirements. To kick off this project, the ADAPT Hand v1 is used to cook a full english break at an ICRA2023 competition.

Developing a tendon driven soft-rigid hand platform with built in compliance through a soft skin, series elastic actuators in finger joints, and an impedance controlled wrist - plans to open source/publish.

Using a 'physical twin' of a raspberry, a sensorized physical simulator, we trained a robot to harvest real raspbrerries. Field tests showed success in promise in this method of field robot development. Ongoing work to improve the twin.

Humans have a vast array of multi-modal sensors, where we use actions to gain more infromation about the world. How can we do the same for robots?

How do we scale up soft robots? We develop a soft robotic shark to a) demonstrate the scale at which we can develop soft robots and b) study how a soft swimming robot can be studied and scaled up (in size) systematically.

In collaboration with the Biodiversity Change group at UNIL, this project aims to develop a 'physical twin' of a flower to automatically capture insects to analyse their pollen content, by mimicking a flower.

Various projects centered around developing soft sensors with tunable geometric and sensing properties. Sensors include 3D printable conductive TPU/piezoelectric TPS, barometric sensor, conductive fabric, hall effect sensing, etc..

We developed a robot inspired by the Feather star, an aquatic animal which can detach its 'feathers' to escape from pray. The robot can equally detach its robotic feathers.

A short project on developing and analysing a dynamic frisbee throwing robot using two BLDC motors and the ODrive controller. The project was motivated through my hobby: playing ultimate frisbee. 

How can robots feel pain and how can that pain signal help the learning of the robot? This robot was built with a 'spinal cord reflex system' and trains a NN to learn from a painful experience. 

Making an controlling multi-fingered robotic hands can be challenging. We developed a robotic hand with flexure joints which can be printed in a single pieces, simplifying fabrication. The control is also simplified through a synergies inspired method: combinatorial actuation to constrain certain joints in parallel, minimizng actuator count.

Master's thesis project on the optimal control of a variable stiffness actuator. We formulate an 'allocator' to treat VSA as a rigid actuator, with an objective function based on stiffness minimization, energy minimization, etc.

An omelette cooking robot was developed to optimize an omelette based on human feedback on the flavour, appearance, and texture. We show the use of batch bayesian optimization yields better optimization performance than sequential bayesian optimization, due to sampling challenges from human evaluators.