- Trener Robotics
- San Jose, CA
- Full-Time
- 4 days ago
Robotics Research Engineer (VLA Engineer).
Before you go
Before you leave us, sign up for our email alerts
We don't do job spam, just the best digital jobs delivered straight to your inbox.
Robotics Research Engineer (VLA Engineer): our view in 3 lines...
- The Role:This role is for an engineer who applies research on vision-language models to robot control for industrial manipulation tasks.
- The Person:The person will own the full loop of data, training, evaluation, and on-robot deployment for manipulation policies, fine-tuning Vision-Language-Action models for industrial tasks like assembly and machine tending.
- Requirements:The ideal candidate has strong Python skills, working proficiency in C++, experience training models in PyTorch or JAX, familiarity with ROS or ROS 2, and comfort with experiment tracking tools like Weights & Biases or MLflow.
About the role
About Trener
Trener is a Physical AI company building software-defined robotics for industrial automation. Our platform is built on the ROS 2 ecosystem, and our product, Acteris, is running in production with real customers on real factory floors — not in a demo lab.
Backed by a $32M Series A, we operate from dual headquarters in San Jose, California and Trondheim, Norway, with teammates across six countries. We work by four values: Move Fast Together, Earn Customer Champions, Deliver with Excellence, and Challenge the Possible.
The role
You will turn VLA research into robot behavior that works on a production line. Industrial work cells are high-mix and unforgiving: parts vary, lighting changes, and a policy that succeeds 80% of the time in simulation is not a product.
As a Robotics Research Engineer in T-Labs, you will own the full loop — data, training, evaluation, and on-robot deployment — for foundation-model-based manipulation policies. You will work alongside our AI compute, perception, and integration engineers to move models from paper to pilot to customer deployment.
What you’ll do
-
Design, train, and fine-tune Vision-Language-Action models for industrial manipulation tasks such as pick-and-place, kitting, machine tending, and assembly.
-
Build and scale robot learning data pipelines: teleoperation and demonstration collection, data curation, augmentation, and synthetic data from simulation.
-
Adapt pretrained vision-language backbones to robot action spaces, including action tokenization, diffusion or flow-matching action heads, and multi-camera inputs.
-
Define evaluation protocols and benchmarks that measure real-world success rate, robustness, and generalization to new parts, cells, and instructions.
-
Close the sim-to-real gap using simulation environments (Isaac Sim/Isaac Lab, MuJoCo, or Gazebo) and domain randomization.
-
Deploy policies on physical robots through ROS 2, working with the AI Compute team to meet latency and memory targets on edge hardware (quantization, distillation, TensorRT).
-
Track and reproduce state-of-the-art research, and share findings through internal tech talks, write-ups, and — where appropriate — publications at venues such as CoRL, RSS, ICRA, and IROS.
-
Partner with Product, Integration, and customer-facing teams to turn customer use cases into learnable tasks and measurable milestones.
What you’ll bring
-
MS or PhD in Robotics, Computer Science, Electrical/Mechanical Engineering, or a related field — or equivalent industry experience.
-
3+ years of hands-on experience in robot learning, imitation learning, or reinforcement learning applied to manipulation.
-
Direct experience training or fine-tuning large multimodal models (VLMs, VLAs, or transformer-based policies) in PyTorch or JAX.
-
Strong Python skills and working proficiency in C++.
-
Experience running learned policies on physical robot hardware, not only in simulation.
-
Familiarity with ROS or ROS 2.
-
Comfort with distributed GPU training and experiment tracking (e.g., Weights & Biases, MLflow).
-
A track record of shipping research into working systems, and clear communication across research and engineering teams.
Nice to have
-
Publications at CoRL, RSS, ICRA, IROS, NeurIPS, ICML, or CVPR on robot learning or embodied AI.
-
Contributions to open-source robot learning projects.
-
Experience with diffusion policies, flow matching, or world models for control.
-
Hands-on work with industrial arms (e.g., UR, ABB, KUKA, FANUC) and grippers, force/torque sensing, or tactile sensing.
-
Experience optimizing models for edge inference (quantization, pruning, distillation, TensorRT, NVIDIA Jetson).
-
Background in large-scale teleoperation or data collection infrastructure.
-
Experience at a robotics startup or in a production manufacturing environment.

