EmergingRobot learning and manipulation

Vision-Language-Action Researcher

Also called VLA Researcher

A Vision-Language-Action Researcher builds models that turn instructions and camera input into safe, bounded robot actions. They matter because they decide whether a robot can follow a plain request in a real, untidy setting.

What they do

They train and adapt multimodal policies that take images and language and produce robot actions. Much of the day goes on choosing action representations, preparing demonstration and simulation data, and running training experiments. They test in simulation first, then on real hardware, and watch closely where the policy fails. They work with controls and hardware engineers on motion planning, limits and safety checks, so the model proposes and the stack constrains. They write evaluations that vary objects, lighting, phrasing and scene layout, and they trace each failure to perception, language grounding or control.

What a strong hire delivers

In the first months, a strong hire gets a working baseline policy running on the team's own robot and in simulation. They set up a repeatable evaluation that shows how well instructions are followed and how behaviour holds up in new settings. They document failure modes, agree safety limits with the hardware team, and give the company a clear view of what to build next.

How you can tell they are good

Ask them to walk through a policy they trained and then tested on a real robot. Look for honest detail on what broke and why. Ask how they measured grounding, as opposed to the robot simply repeating a familiar motion. Strong people explain their action representation choices and their safety limits. Be wary of simulation-only results and vague claims about generalisation.

Why it is hard to fill

Almost nobody carries this title yet. The work sits inside robotics, ML research or controls jobs, so titles tell you little. Candidates tend to be strong on one side: either model training or hardware and control. Few have done both and put a policy on a real robot. Searches by keyword mostly return the wrong people, so judging real evidence takes effort.

Where to find them

Look at robot learning researchers, manipulation and perception engineers, and ML researchers who moved toward embodied work. Controls and mechatronics engineers who learned modern multimodal models are another source. They sit in university robotics labs, frontier lab robotics teams, humanoid and manipulation start-ups, and automation companies building general-purpose robots. Open-source robotics contributors are worth checking too.

Indicative pay

Not enough published evidence yet to give a fair range for this role, which is common for work this new. We benchmark it for you as part of a search.

Interview questions for a Vision-Language-Action Researcher

  1. Describe a language-conditioned policy you took onto real hardware. What failed first?

    A strong answer names concrete failures such as perception errors, ambiguous instructions or contact problems, and what they changed. Worry if the story is only about simulation, or if every failure is blamed on data volume.

  2. How did you choose your action representation, and what trade-offs did you see?

    Good answers compare options such as joint targets, end-effector deltas or discretised tokens, covering smoothness, latency and precision. Worry if they used a default without being able to say why.

4 more questions, with what a strong answer covers, are in the Role Kit.

Job description you can post

We are building robots that follow natural instructions and act safely in real environments. We are looking for a Vision-Language-Action Researcher to train and evaluate multimodal policies, take them from simulation onto hardware, and help us understand where they work and where they do not.

What you will do

  • Train and adapt vision-language-action policies for manipulation tasks
  • Design and compare action representations
  • Build evaluations for instruction grounding and generalisation across settings
  • Validate policies in simulation and on real robots
  • Work with controls engineers on motion planning and safety limits
  • Analyse failures and feed findings back into data and model choices
  • Document methods so the team can reproduce results

What you will bring

  • Demonstrated work training policies that use both vision and language
  • Experience running learned policies on real robot hardware
  • Solid grounding in robotics and control
  • Practical computer vision and reinforcement or imitation learning experience
  • Comfort with simulation tools and ROS-based systems
  • Clear written analysis of experiments and failures

Nice to have

  • Experience with motion-planning stacks
  • Work on sim-to-real transfer
  • Open-source robotics contributions

Free to use and adapt. If it saves you time, a link back to this page is appreciated.

Skills that matter

  • Robotics and controlAdvancedPerceive, plan and act through physical systems
  • Reinforcement learningAdvancedOptimise policies from interaction and reward
  • Simulation and digital twinsWorkingBuild useful virtual representations of environments
  • Computer visionWorkingAnalyse or generate visual information

Other roles in robot learning and manipulation

Hiring nowRobot Learning EngineerA Robot Learning Engineer trains policies that let robots complete defined physical tasks reliably. The work turns a promising demo into behaviour a…Hiring nowRobot Manipulation ResearcherA Robot Manipulation Researcher builds the grasping and object-interaction methods that let robots handle real things reliably. They turn lab demos…

Related roles in other families

Hiring nowAutonomy Perception EngineerAutonomous navigation and field roboticsAn Autonomy Perception Engineer turns raw sensor data into a reliable picture of the world that a machine can act on. It decides whether an…Hiring nowLocalisation and Mapping EngineerAutonomous navigation and field roboticsA Localisation and Mapping Engineer builds the system that tells a robot or vehicle where it is and what surrounds it. Without that, nothing…Hiring nowMotion Planning EngineerAutonomous navigation and field roboticsA motion planning engineer designs the software that decides how a robot or vehicle moves, so it reaches its goal without breaking physical limits…Hiring nowField Robotics Integration EngineerAutonomous navigation and field roboticsA Field Robotics Integration Engineer gets autonomy software and sensors working together on a real machine outdoors. It matters because autonomy…

How we wrote this

Role definitions are our own editorial work, checked against occupational classifications, standards, research and employer pages. Titles vary by company, and this is a description of the work, not a job advert.

← Robot learning and manipulationAll new job families

Start a search

Hiring a Vision-Language-Action Researcher?

Tell us the work. We find the people already doing it, whatever their job title says, and you meet them already interviewed for your role.