Applied AI Scientist, Simulation & Reinforcement Learning

At Locus Robotics, we build AI-powered systems and intelligent robots that keep global supply chains running. Our platform combines advanced AI, real-time decision-making, and autonomous robotics to help leading companies improve efficiency, scale operations, and adapt to constant change. Locus Robotics is a place to do meaningful work with real-world impact, where your ideas move quickly from concept to deployment. We invest in our people, encourage continuous learning, and give you the opportunity to grow your career while building technology that is used every day at global scale. As an Applied AI Scientist, Simulation & Reinforcement Learning, you'll own the simulator behind SDL, our programme building a world-first, reinforcement-learning-driven orchestration framework for people and robots. That framework needs a simulator that is trustworthy enough to develop, tune and validate it before it touches a live warehouse. You will take over day-to-day ownership of our discrete-event simulation (built on SimPy), make it measurably closer to reality, and use it as the training and evaluation ground for learning-based policies. Simulation comes first; reinforcement learning grows to be the larger part of the role as the simulator matures. You'll compare your simulations against live operational data from some of the world's largest fulfilment sites, and your work will decide what is validated and shipped to fleets of robots in customer sites worldwide. This is not a notebook role. The simulator you own is the foundation that decides what ships to customer sites worldwide. You will work in a small, collaborative team where research meets production, as part of Applied Data Science (ADS), reporting to Oscar Mendez. • Location: This is a remote position based in the United Kingdom. Candidates must reside in the UK. Right to Work: Candidates must have the legal right to work in the UK. Locus Robotics is unable to provide visa sponsorship for this position, now or in the future. Responsibilities Own the simulator • Own and develop our SimPy-based discrete-event simulator, including its logic, design choices and the fragile areas that need care. • Pay down technical debt: improve test coverage, code quality, reproducibility and developer tooling. • Build the experiment infrastructure around the simulator: orchestration, study management, reporting and automated runs across benchmark sites. • Generate and maintain site data (e.g. distance files, configs) needed to simulate customer warehouses. • Support scenario testing and sensitivity analysis for our order-engine and orchestration teams. Tune and close the sim2real gap • Define, measure and reduce the sim-to-real gap against real site data. • Run parameter optimisation and tuning studies (e.g. Optuna), and diagnose anomalies and non-determinism. • Model warehouse processes such as congestion, picker behaviour and queueing at the right level of fidelity. • Communicate results and confidence clearly to engineering, product and operations stakeholders. • Use Optimization, RL & DL to tune the simulator Grow into reinforcement learning • Turn the simulator into a proper RL environment (Gymnasium/Gym-style interfaces) with fast, repeatable, scalable training runs. • Design, train and evaluate RL policies, starting with guidance generation, and benchmark them against current heuristic algorithms. • Guard against overfitting to simulator artefacts, and design experiments that show what transfers to the real world. • Prototype, evaluate and productionise learning-based components alongside the wider ML, MLOps and OE teams. • Stay current with RL, multi-agent systems and simulation research, and assess what is applicable to Locus. • Contribute to internal technical paper