Research
Robot Talk’s latest episode features a conversation with Dexter Shepherd of the University of Sussex on applying artificial intelligence to help robots understand and adapt to their environments. The discussion centers on machine learning for bio-inspired tactile sensing and robotic locomotion — two capabilities that let machines interpret physical contact and adjust how they move.
Shepherd is a Postdoctoral Researcher at the University of Sussex, where his work spans machine learning for bio-inspired tactile sensing and robotic locomotion. According to the episode description, his research combines embedded AI systems, spatial-temporal models, and reinforcement learning, with the aim of enabling robots to classify textures, orient themselves, and adapt their movements.
The approach brings together three technical threads: embedded AI systems that run computation on-device, spatial-temporal models suited to processing contact signals over time, and reinforcement learning for adaptive control. The stated goals — texture classification, self-orientation, and movement adaptation — point to a sensing-and-control loop in which tactile information informs locomotion. The source does not report specific experimental platforms, datasets, quantitative results, or a publication venue for this work.
His work combines embedded AI systems, spatial-temporal models, and reinforcement learning to enable robots to classify textures, orient themselves, and adapt their movements.
Beyond the research itself, Shepherd leads and coordinates AI training initiatives in Africa focused on making electronics and applying AI on embedded devices — an outreach dimension that connects his embedded-systems expertise to capacity-building efforts.
Robot Talk is a weekly podcast exploring robotics, artificial intelligence, and autonomous machines. The episode is presented as a discussion rather than a technical paper, so listeners seeking detailed benchmarks or hardware specifications will not find them in the episode summary; the value here is an overview of how bio-inspired tactile sensing and embedded learning are being framed as a route to more adaptive robot behavior.
Source: Robohub