Research Scientist, Reinforcement Learning

We are hiring a Research Scientist to develop reinforcement-learning methods for embodied intelligence. The objective is to help foundational models acquire new capabilities through interaction, improve with feedback and adapt their behaviour when a task changes.

You will work across the learning algorithm, reward signal, training environment and resulting policy. Simulation is central to this role: it provides a way to study exploration, scale experience collection and test behaviour before physical evaluation. We are looking for someone who can turn an open research question into a training system and explain what the learned policy is actually doing.

Responsibilities

• Develop reinforcement-learning algorithms and training objectives for policies that must learn from demonstrations, feedback and interaction.
• Investigate reward design, exploration, credit assignment and curricula for tasks that require sustained sequences of actions.
• Build experiments that connect pretrained foundational models to policy learning and adaptation.
• Use simulation to study sample efficiency, robustness and transfer, including the effects of environment design and training distributions.
• Diagnose training instability, reward exploitation and policy failures through targeted experiments and trajectory analysis.
• Collaborate with robotics and systems engineers on experience collection, distributed training and evaluation on physical tasks.

Qualifications

• BS, MS or PhD in machine learning, robotics, computer science or a related quantitative field, or demonstrated work of equivalent technical depth.
• Original reinforcement-learning research or substantial learning systems, with hands-on responsibility for training policies and evaluating their behaviour.
• A research record demonstrated by leading contributions at venues such as NeurIPS, ICML, ICLR, CoRL or RSS, or research systems that establish comparable algorithmic and experimental depth.
• Practical experience implementing RL algorithms, managing rollout collection and debugging optimisation, rewards and environment behaviour.
• Strong Python and PyTorch skills, supported by a sound understanding of probability, optimisation and sequential decision-making.
• The ability to design controlled evaluations and distinguish a policy improvement from an easier environment, altered reward or change in available data.

Preferred experience

• Offline or online RL, imitation learning, policy optimisation or learning from human feedback.
• Physics simulation, manipulation, domain randomisation or transfer between simulated and physical systems.
• Distributed rollout systems, curriculum learning or adaptation of pretrained models through reinforcement learning.

Application materials

Include your CV and selected research or engineering contributions. Describe an RL training project you personally developed: the environment, learning signal, policy, principal failure modes and final result. Include code, publications or rollout demonstrations where available, and explain the next problem you would want to investigate.