Senior Machine Learning Engineer – VLA

Bonsai Robotics • San Jose, CA

8h

About The Position

We're looking for a Machine Learning Engineer who can own the full lifecycle of training and deploying end-to-end Vision-Language-Action (VLA) models for outdoor autonomy. You'll build the models that allow any vehicle—from our Amiga platform to heavy off-road equipment—to navigate, act, and adapt in unstructured outdoor environments using raw sensor inputs (cameras, IMU, GPS, LiDAR, and more). We have collected a large dataset of heavy equipment working in the field from deployments and are looking for people who can leverage massive real-world data as well as data from targeted data collection to train a VLA model. This is a high-impact, end-to-end role: you'll touch everything from data pipelines and model architecture to real-world deployment on edge hardware.

Requirements

• 3+ years applying deep learning to real-world robotics or embodied AI problems using PyTorch, JAX, Ray, or similar frameworks
• 3+ years building, deploying, and maintaining ML models in production—not just research prototypes
• Strong practical experience with behavior cloning, reinforcement learning, or other data-driven control approaches
• Familiarity with multimodal model architectures (vision-language models, VLAs, or similar)
• Comfort working across the stack: data infrastructure, model training, optimization, and on-device deployment
• Experience working with real sensor data (camera, IMU, GPS, LiDAR) in noisy, unstructured environments

Nice To Haves

• Experience with flow-matching policies, action-chunking transformers, or other recent advances in learned manipulation/navigation policies
• Hands-on work with TensorRT, ONNX, or other model optimization toolchains for edge deployment
• Published work at ICRA, IROS, CoRL, CVPR, NeurIPS, ICML, or similar venues
• Experience with ROS2 or robotics middleware in production systems
• Background in agriculture, construction, mining, or other outdoor/off-road domains

Responsibilities

• Design, train, and iterate on VLA and other learned behavior policy architectures for vehicle control in diverse outdoor environments
• Build and maintain robust data pipelines—ingestion, curation, labeling, and versioning—to support reproducible, high-quality training at scale
• Develop evaluation frameworks: offline metrics, simulation-based testing, and real-world field validation loops
• Optimize models for deployment on edge compute (NVIDIA Jetson and similar), addressing latency, memory, and throughput constraints
• Collaborate closely with perception, controls, platform, and field operations teams to integrate learned policies into our full autonomy stack
• Instrument and monitor deployed models in production, diagnosing failure modes and feeding insights back into the training loop
• Stay current with the rapidly evolving landscape of foundation models for robotics and bring new ideas from research into practice

What This Job Offers

• Job Type: Full-time
• Career Level: Senior
• Education Level: No Education Listed