AI & Robotics Research Engineer - Learned Dexterous Manipulation

Overview

Proception is seeking an expert in reinforcement learning to help build foundation models for dexterous humanoid manipulation. You will lead the design of scalable RL pipelines, train robust policies on high-DOF robotic hands, and deploy intelligent behaviors that push the limits of real-world robotic control.

• Design and deploy scalable RL systems for high-DOF humanoid and dexterous robotic hands
• Develop foundation models for manipulation tasks such as grasping, in-hand reorientation, and tool use
• Train policies that are robust to real-world uncertainty, including contact variation, sensor noise, and hardware drift
• Build and maintain physics-accurate simulation environments using Mu Jo Co  or Isaac Lab
• Apply advanced RL and control techniques to model complex object–hand interactions
• Develop imitation and hybrid learning pipelines from human demonstrations
• Leverage multimodal feedback (vision, proprioception, tactile sensing) for adaptive manipulation
• Explore learning-driven optimization of robot hardware, morphology, and actuation strategies
• Work closely with hardware, perception, and systems teams to deploy policies on real robots

• MS or PhD in Robotics, Computer Science, Machine Learning, or equivalent industry experience
• Deep expertise in reinforcement learning, policy optimization, and continuous-control systems
• Strong foundation in probability, optimization, and linear algebra
• Hands-on experience training and deploying RL policies on real robotic systems
• Proficiency in Python and C++ on Linux/Unix platforms
• Experience with robotics simulators such as Mu Jo Co  or Isaac Sim
• Familiarity with imitation learning, inverse RL, or hybrid learning approaches
• Ability to design scalable training infrastructure and manage large-scale experiments
• Experience with tactile sensing, contact-rich manipulation, or robot morphology optimization is a plus
• Self-driven, curious, and excited to build systems that work beyond simulation