Model Training & Deployment: Design, train, and deploy deep learning models for semantic reasoning, specifically tailored to achieve the extended spatial range and high fidelity required for high-speed highway environments.

Cross-Functional Collaboration: Collaborate with the Scene Intelligence, Semantic Grounding, and PCP Mapping teams to adapt and elevate the unified machine learning stack for highway scenarios.

Requirements & Validation: Partner with downstream motion planning teams to define semantic representation requirements, establish robust validation workflows, and ensure model outputs meet strict safety and clearance metrics.

Optimization: Optimize deep learning models for real-time inference efficiency, ensuring low-latency execution within the rigorous compute constraints of the Zoox vehicle platform.

Edge Case Resolution: Investigate and resolve perception-related regressions and edge cases found in high-speed driving simulations and live fleet data.

Strategic Architecture: Contribute to the long-term "North Star" architecture for Perception Semantic Reasoning, paving the way for scalable fleet deployment across new vehicle platforms.

MS (3–5 years) or PhD (0–2 years) in Computer Science, Robotics, Electrical Engineering, or a related field, with professional software engineering experience — ideally in autonomous driving, robotics, or computer vision.

Deep understanding of 2D/3D computer vision, semantic segmentation, and deep learning architectures.

Exceptional programming skills in modern C++ and Python.

Hands-on experience with modern deep learning frameworks like JAX or PyTorch.

Proven track record of deploying real-time machine learning models on resource-constrained embedded systems or on-bot hardware.

Prior experience dealing with highway autonomous driving scenarios and their specific mapping/perception challenges.

Familiarity with state-of-the-art, BEV, Sparse Transformer architectures and Vision-Language Models (VLMs).

Strong publication record in top AI conferences or journals (e.g., CVPR, ICCV, ECCV, ICML, NeurIPS).