Skip to main content

Xdof

Member of Technical Staff, Embodied Capabilities

  • Confirmed live in the last 24 hours
  • $150k–$280k
  • Mid level
  • Full-time
  • Remote · San Mateo Hybrid
  • Added today

About this role

At XDOF, we’re at an inflection point. Frontier labs are racing to build general-purpose robots, and high-quality training data is the bottleneck. We’re building the foundation behind the foundation models – the data collection systems, operational capability, exabyte-scale data warehouse, and model evaluation platform – to help our partners drive the field forward.
We’re looking for an Embodied Capabilities researcher to figure out what frontier models can actually do in the physical world, and where they fall short. You’ll design the tasks behind our benchmarks and RL environments, prove them out on our real robots, and use our data to decide what to build next. Examples of things you might work on: designing a benchmark suite that separates today’s best VLAs, finding the tasks where LLM agents break down, and turning data insights into new tasks and targeted data collections.
What You’ll Do
Embodied Capabilities researchers own our tasks end to end: what we measure, how we grade it, and whether it’s a good measurement. Sample projects include:

  • designing tasks for LLMs, VLAs, and WAMs that isolate real robotics capabilities, from precise contact to long-horizon planning and recovery from mistakes

  • implementing tasks on our evaluation platform, including reward functions, verifiers, success criteria, and environment configs

  • validating tasks on real hardware with our operators to prove they’re possible, prevent reward hacking, and calibrate their difficulty

  • building the simulated versions of tasks for our RL environments, including measuring real-world variation, like the spread of starting positions

  • owning the task taxonomy that decides which capabilities each task tests and which environments it belongs in

  • analyzing evaluation results and our collected data to find capability gaps, unreliable tasks, and coverage gaps

  • publishing findings on where frontier models stand in the physical world

You might be a good fit if you have

  • strong programming skills in Python

  • hands-on experience with robot manipulation, and an intuition for what makes a physical task hard

  • strong data analysis skills, including working with large episode datasets and video

  • a solid grasp of experimental design and statistics, and thought hard about what a score actually measures

  • a working knowledge of current VLAs and LLM agents, and what they can and can’t do

  • a creative, adversarial mindset: you enjoy designing challenges and finding ways around them

Bonus points if you have

  • designed robotics competitions or benchmarks (e.g. RoboCup, BEHAVIOR, RoboArena, IROS or ICRA challenges)

  • experience with LLM-as-judge, reward modeling, or RL post-training

  • worked with simulation (MuJoCo, Isaac Sim, or similar) or real2sim pipelines

  • fine-tuned a VLA or run model baselines yourself

  • experience with CAD or 3D printing for fixtures and repeatable setups

Written by Xdof. Original job post

Skills mentioned

  • Data Analysis
  • Data Warehouse
  • Python
Apply with Autofill

Opens the application — the Jobs AI extension fills it for you. Set up autofill

Opens the official application on the employer’s site. No login required.

Xdof

View all jobs at Xdof

Preparing likely interview questions for this role…