Academic research
Optoelectronic tactile sensing makes contact observable. Learning turns those observations into useful state estimates. Control brings them back into the physical world. Explore this path through published work by Professor Nathan Lepora and collaborators.
Sense. Learn. Manipulate.
Nathan Lepora · Professor of Robotics & AI · University of Bristol
A connected research path
These themes connect the physical design of a soft fingertip with the models and feedback that guide a robot’s actions.
Optoelectronic tactile sensing
The TacTip family uses compliant skins and internal markers to translate contact deformation into image measurements. Sensor shape and mechanics become part of the sensing system.
TacTip family · 2018
Tactile learning
Deep learning estimates contact pose from tactile observations. Tactile Gym then connects simulation, reinforcement learning and real sensors for tasks such as pushing and surface following.
Optimal Deep Learning · 2020 / Tactile Gym 2.0 · 2022
Dexterous manipulation
Pose and shear guide continuous tactile control. Bimanual manipulation, in-hand rotation and multi-finger grasp control extend this feedback from a single contact to coordinated action.
Bi-Touch · 2023 / AnyRotate · CoRL 2024 / Shear grasp control · 2025
Selected publications
A selection of published work spanning sensor design, tactile learning and robot control. Follow each link to the original publication.
A historical view of robotic touch, tracing recurring challenges and the opportunities ahead for tactile robotics.
Read the paperMulti-finger tactile feedback regulates grasping during delicate-object handling, pouring and human-guided movement.
Read the paperA tactile policy trained in simulation transfers to a real hand for object rotation across axes and hand orientations.
Read the paperContact pose and shear estimates support smooth tracking, surface following and single- or dual-arm object pushing.
Read the paperTwo tactile robot arms learn coordinated pushing, reorientation and gathering, with policies transferred from simulation to hardware.
Read the paperA shared simulation-to-real approach compares TacTip, DIGIT and DigiTac on pushing, edge following and surface following.
Read the paperDeep-learning methods estimate the contact pose of surfaces and edges while accounting for variation such as sliding-induced shear.
Read the paperA family of soft, 3D-printed tactile sensors uses biomimetic structures to transform skin deformation into internal marker motion.
Read the paperAnyRotate was presented at CoRL 2024; its PMLR proceedings record was published in 2025. Paper titles retain their original published terminology.
Browse the complete research archiveExplore the original research
This page follows a selection of Nathan Lepora and collaborators’ published research. Visit the personal website for the broader programme, demonstrations and full publication archive.