Robot bodies and brains co-evolving on AWS
Evolutionary robotics needs massive parallel simulation: thousands of generations, dozens of robot designs, each one rebuilt in physics from scratch. The lab pipeline ran on local hardware that could only evaluate one population at a time, and class projects had to queue. The science was throttled by the infrastructure, and so was the teaching.
Re-architected Robogen as two cooperating services (an evolution engine for population, selection, mutation, reproduction; and a physics simulator for fitness evaluation) talking over the network so the simulator could scale independently. Migrated the simulator onto AWS with autoscaling, peaking at 15 EC2 instances during class demand. Secured AWS Cloud Credits for Research to fund the compute. Taught the platform as TA on the EPFL Evolutionary Robotics course, supervising 20+ master's students through end-to-end evolutionary experiments.
100+ master's students used Robogen for class projects. Dozens of robot morphologies evolved across thousands of generations, with the best designs printable on the lab's 3D printer and wired up with off-the-shelf electronics. Featured on the AWS Public Sector blog as a reference case for autoscaling research compute. The optimisation, search, and embodied-learning intuitions from this work underpin every agentic AI system I design today.
Research-tier work like this is rarely a fixed-scope engagement. The closest commercial shape is an AI Transformation Sprint to scope and prove the experiment, then a custom-quoted full-system build to migrate the simulation pipeline to cloud-scale parallel runs.
Behind on AI and not sure where to start?
Book a 30-minute call. I'll tell you what's worth building, what isn't, and whether I'm the right person for it.