Research Scientist, Robot Learning

Good AI Labs is hiring a Research Scientist to advance learning for robotic manipulation. You will develop policies that acquire skills from demonstrations and interaction, then investigate how those skills transfer to new objects, task variations and operating conditions.

The work connects foundational models with imitation learning, reinforcement learning and simulation. You will own the training and evaluation of a research direction, working with engineers to understand how a policy behaves when it controls a robot. We are interested in researchers who can move between a new learning idea and the practical details that determine whether it works.

Responsibilities

• Design and train policies that use visual observations, task context and interaction history to perform manipulation tasks.
• Develop imitation-learning and reinforcement-learning methods that improve skill acquisition, adaptation and recovery.
• Investigate how demonstration quality, data mixtures and action representations affect learning and generalisation.
• Build simulation experiments and evaluation protocols that test task transfer and expose gaps before physical deployment.
• Work with robotics engineers to analyse failed rollouts, sensing delays, action timing and differences between simulation and hardware.
• Establish the relationship between task performance, training data and inference requirements, and use it to guide further model development.

Qualifications

• BS, MS or PhD in robotics, machine learning or a related discipline, or demonstrated research and robotics work of equivalent depth.
• A strong record in robot learning, with direct experience implementing, training and evaluating policies for embodied tasks.
• Original contributions demonstrated through publications at venues such as CoRL, RSS, ICRA, IROS, NeurIPS or ICML, or substantial robot-learning systems with comparable technical merit.
• Strong Python and deep-learning implementation skills, including the ability to diagnose training failures from data, model behaviour and rollouts.
• A working command of robot observations, action spaces, control frequency and closed-loop evaluation.
• Evidence that you can lead a research project through implementation, experimentation and a result that holds beyond its easiest test cases.

Preferred experience

• Foundational models for robotics, multimodal policies, multi-task learning or learning from heterogeneous demonstrations.
• Simulation, reinforcement learning, domain randomisation or sim-to-real transfer.
• Physical manipulation experiments, data collection or adapting policies across robot embodiments.

Application materials

Include your CV and selected papers, code or demonstrations. Describe a policy you trained, the data and environment it used, your contribution and the conditions under which it succeeded or failed. We welcome both academic research and substantial engineering work on learning-based robots.