Elvis Nava, co-founder and CTO of Mimic Robotics, discusses dexterous robot hands and learning-based manipulation with host Devang on The Humanoid Hub.
Nava says Mimic chose dexterous hands because its team could build advanced hardware and because human-like dexterity can help align robot learning with human data. The company is focusing on stationary industrial tasks and is avoiding legged locomotion for now, which he considers a separate source of complexity.
Nava describes tendon drives as a way to make a hand in a human-like shape and size while supporting back-drivability and thermal performance. Mimic uses a larger motor block for strength and reliability, and Nava says proprioceptive sensing is important for controlling the hand.
Discussing an Audi demo that fits flexible rubber around a door frame, Nava points to on-site data collection, end-to-end learning, policy design, and reliable hardware. He says the demo uses vision and robot position sensing, with no tactile sensing in that version.
Nava explains that video models can learn motion and dynamics during pretraining, potentially making better use of human and cross-embodiment video than a vision-language model. In MimicVideo experiments, a Cosmos backbone fine-tuned on about 200 hours of robot data was paired with an action decoder trained on 400 task episodes. Nava reports roughly ten times greater data efficiency than fine-tuning π0.5 on the tested tasks, and says the approach can run in real time by decoding actions from video-model representations.
Nava wants Mimic to expand across industrial manipulation, beginning with repeatable tasks and then moving toward more general-purpose work. He expects open-ended home robotics to need more reasoning and planning, while physical deployment will take time because robots must be built and delivered.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.
Scott Walter, a robotics engineer and advisor, joins Over The Horizon hosts and guests to discuss Sharpa’s Dairy Queen deployment, Mimic’s dexterous hand, and Sunday Robotics’ home laundry results.
Elvis, co-founder and CTO of Mimic Robotics, discusses video-action models and robot learning with the RoboPapers hosts.
