VLA Engineer

Role Overview

The WoRV team develops Mobile Manipulation VLA that performs both locomotion and manipulation through a single model in environments not originally designed for robots — such as farmland, factories, and construction sites.

This role defines customer problems from the field as technical challenges and turns them into Safe, Reliable, and Scalable products.

We work full-stack from simulation and hardware up to VLA / End-to-End autonomous driving models. Rather than decomposing capabilities into separate modules, we train a single unified model. This is not a one-off project — we build product capabilities that can be repeatedly applied across multiple customers. As new tasks and field sites come in, you will directly design and train the intelligence of a robot foundation model that bridges navigation and manipulation across diverse environments and platforms.

Responsibilities

• Design VLA models integrating driving and attachment control, and co-define safety requirements and evaluation criteria.
• Propose and implement training pipelines combining Pre-training, Imitation Learning, RL Post-training, Sim2Real, and other techniques.
• Measure and train robot foundation model intelligence through Closed-loop (simulation) and Open-loop (dataset benchmark) evaluation pipelines, and validate in real field environments.
• Analyze real-world logs and failure cases to establish improvement roadmaps including model architecture and training data priorities.
• Collaborate with PM / Control / Simulation / Data teams to connect research to real services.

Qualifications

• Broad understanding of AI/ML with deep expertise in at least one area
• Strong C++ / Python implementation skills with real-world system development and experimentation experience
• End-to-end experience connecting at least one problem from experiment to deployment to improvement
• Experience reflecting inference latency, model size, memory, and safety constraints of real-vehicle deployment into model and training decisions
• Ability to examine model outputs alongside field failure cases and structurally define when and how a model fails
• Experience running fast cycles from paper to code to real-vehicle deployment

Preferred Qualifications

• Awards in science, mathematics, or physics olympiads
• Experience deploying models or systems on real equipment, vehicles, or robots
• Strong research track record in machine learning or robotics (publications in top venues)
• Experience with Robotics / Autonomous Driving / Physical AI / VLA / RL / Imitation Learning
• Experience with Motion Planning / SLAM / Localization / Control Integration
• Experience with ROS2, C++, edge inference, or on-device optimization
• Experience with brownfield environments, noisy sensors, partial observability, and edge cases
• Experience with safety validation, fallback logic, or evaluation harnesses

Who We're Looking For

• Someone who asks "which technology fits this problem?" before anything else
• Someone who knows benchmark accuracy and real-field performance are different, and is interested in closing that gap
• Someone who can design for both immediate customer value and future scalability
• Someone who wants to build reusable product capabilities, not just local optima