Robot Talk · 25 Apr 2025 · 36m · Episode 118
Miranda Lowther, a PhD researcher at the University of Bristol, discusses her development of a sensorized electronic skin (e-skin) for prosthetic liners. She explains how a network of saline-filled channels changes resistance under pressure, allowing machine learning to map discomfort and improve prosthetic fitting. Lowther also explores morphological computation, where the physical structure of the device contributes to its intelligence, and addresses challenges in soft robotics such as material predictability and interdisciplinary collaboration.
Where this conversation describes AI being used to build or deliver a product. Part of the AI in practice tracker.
Engineering & simulation
Machine learning was used to map resistance changes from the e-skin channels to specific locations and sizes of forces, enabling the quantification of pressure points on a prosthetic limb.
“so then what we do is that we collect all these readings and use machine learning to then map these resistance changes or impedance changes. To locations and sizes of forces.”
Speaker 2 · Hear it at 6:36
early-stage companies
Other shows on the same ground, closest first.
Professor Gordon Cheng of the Technical University of Munich explains his development of event-driven e-skins for humanoid robots, using off-the-shelf sensors to create a low-power, autonomous touch system that improves safety and enables brain-machine interface applications. He discusses the challenges of scaling sensor data, the move from conventional control to neuromorphic event-based processing, and the potential for sensory substitution in prosthetics. The episode also features commentary from Dr. Giulia D’Angelo and Professor Ralph Etienne-Cummings on the definition of neuromorphic engineering and the current state of the industry.
Professor Gordon Cheng of the Technical University of Munich explains his development of event-driven e-skin for humanoid robots, using off-the-shelf sensors and neuromorphic communication to enable safe, autonomous physical interaction. He details how hexagonal sensor patches covering the robot's body transmit only change-based events to avoid data overload, allowing the robot to sense proximity, force, temperature, and vibration. The discussion with Dr. Giulia D’Angelo and Professor Ralph Etienne-Cummings explores the distinction between device-level and system-level neuromorphic engineering, the challenges of scaling such technology, and its applications in brain-machine interfaces and prosthetics.
Hosts Daniel and Forbode discuss two recent engineering breakthroughs inspired by biology. First, they cover research from the University of Michigan on 'RAILS' (Robust Adversarial Immune-inspired Learning System), which uses B-cell immune system logic to make AI image classifiers robust against adversarial attacks. Second, they analyze electronic skin developed by Pohang University of Science and Technology (POSTECH) and UNIST, which mimics human nerve pathways to achieve low-latency tactile sensing for robotics and prosthetics.
The hosts discuss a new stretchy electronic fiber developed by a team at EPFL that combines high conductivity with extreme flexibility using liquid metal and rubber. They explain the thermal drawing process used to manufacture the fiber, comparing it to making fiber optics or hard candy, and highlight its high gauge factor and durability. The conversation explores applications in smart wearables like knee braces, soft prosthetics, and robotic skins for humanoid robots such as Tesla's Optimus.
Briana Morey, an electrical engineer at MC10, discusses the company's stretchable wearable sensors, including the L'Oreal My UV Patch and the BioStamp RC. She explains the proprietary accordion-like geometry that allows electronics to stretch with skin, the use of NFC for the UV patch, and the BLE connectivity of the BioStamp RC. Morey also shares her background designing high-power electronics for lasers at JPSA and her personal hobby of building Tesla coils.
The summary, chapters and topics are generated from the episode’s audio by this directory, not written by Robot Talk. The episode itself belongs to its publisher.