Senior Robot Learning Engineer

About the role

ABOUT LIGHTSPEED

LightSpeed Build Technologies is revolutionizing the construction industry through AI-powered robotics. Our flagship systems—BRUTE for automated wall panel manufacturing and DEX for on-site collaborative construction—are addressing the global housing crisis by delivering unprecedented speed, precision, and affordability in homebuilding.

POSITION OVERVIEW

As Senior Robot Learning Engineer, you will own how LightSpeed's robots learn and perform physical skills. You will build the data pipeline that captures how people do the work, the policies that learn from it, and the executor that runs every action on the robot, classical or learned, behind one interface.

Our robots do contact-rich, high-variance work: taping seams, routing cable, finishing drywall, setting fixtures. Your systems decide whether a new trade takes days or weeks to teach, and whether it runs reliably once it does.

WHAT YOU'LL WORK ON

DATA AND TELEOPERATION

- Define how we capture demonstrations: teleoperation rigs (leader arms, VR) and egocentric capture of people doing the work

- Own the data pipeline: synchronized observations and actions, episode labelling, dataset versioning and quality checks

- Decide what data each new task needs, and how failures are labelled and fed back

- Partner with the world sensing principal on observation spaces and sensor streams

POLICIES AND TRAINING

- Set the policy architecture: a shared vision and force backbone with task-conditioned heads, so a new skill needs a few demonstrations, not a new model

- Train with imitation learning, diffusion or transformer policies and vision-language-action models; use reinforcement learning where the reward can be clearly defined

- Build the training and evaluation loop, in simulation and on real robots, with held-out evaluations that gate deployment

- Solve manipulation of deformable materials: cable, tape, membrane

- Set coverage and first-pass yield targets per skill, and measure against them

EXECUTION AND CONTROL

- Own the actuate executor: classical control and learned policies behind one contract, so the planner can choose either for each step

- Build the compliance layer for contact-rich work: nailing, pressing, taping, mudding, insertion, with force and impedance control

- Build recovery behaviours for the common failures of each primitive

- Deploy policies on the robot within real-time and safety limits

- Define what each tool must sense and report, so its actions can be learned from and controlled

FOUNDATION AND SKILL PIPELINE

- Co-own the foundation interfaces with the world sensing and agentic planning principals, and own the data and evaluation interface

- For each new trade: collect demonstrations, train and evaluate the policy, add recovery