Intelligent Dexterous Embodied AI Lab · UNC-Chapel Hill
We are a full-stack robot learning lab building intelligent robots that interact with the physical world as naturally and dexterously as humans — from humanoids and dexterous hands to hardware and large models.
Research
IDEAL Lab is a full-stack robot learning lab: we design the hardware, learn the skills, and scale them with foundation models — so ideas travel all the way from first principles to real robots.
The hand is where intelligence meets the world. We work on contact-rich manipulation with multi-fingered hands — grasping, in-hand reorientation, and tool use guided by touch — so robots can handle the long tail of everyday objects, not just the easy ones.
Robots shaped like us can work in spaces built for us. We study whole-body loco-manipulation — walking, balancing, reaching, and carrying at the same time — with unified policies that treat locomotion and manipulation as one problem rather than two, and transfer them from simulation to hardware.
Foundation models bring open-world knowledge; robots give it hands. We build vision-language-action models, world models, and embodied agents with physical reasoning — policies that generalize across tasks, objects, and even robot embodiments.
Learning is only as good as the body it runs on. We design mechanisms, actuation, and control for capable and affordable platforms, and build real-to-sim-to-real pipelines that keep our simulators honest — so what works in simulation works on the robot.
