Physical AI Architect

Job description

**About this role:** Pickle is on the hunt for a dynamic and driven Physical AI Architect to revolutionize the future of warehouse automation. This is a senior technical role for someone who is equal parts deep practitioner and pragmatic builder — someone who understands the theoretical underpinnings of diffusion-based models and optimal control, and who has the track record to prove they can ship these systems into production hardware. You will serve as the technical authority on how modern AI approaches translate into real robot behavior, bridging cutting-edge methods with the reliability and performance demands of high-throughput logistics. If you are energized by closing the gap between research and reality, and you measure success in deployed systems rather than papers, this role is for you. ### Responsibilities: - Serve as the technical architect for Pickle Robot's Physical AI stack, owning the end-to-end design of perception, planning, and control systems deployed on production hardware. - Lead the application of diffusion-based policy learning and optimal control techniques to robot manipulation and picking tasks, with a focus on real-world reliability and cycle time performance. - Drive hardware integration efforts across sensors, compute, and actuators — ensuring AI systems are co-designed with the physical platform from the ground up. - Define the technical roadmap for how diffusion models and optimal control complement each other in Pickle Robot's autonomy architecture, and build internal alignment around that vision. - Partner with firmware, mechanical, and software engineering teams to ensure AI design decisions are grounded in hardware constraints and operational realities. - Identify and resolve performance bottlenecks at the intersection of model inference, motion execution, and hardware throughput. - Mentor senior engineers and help grow the technical depth of the broader autonomy team. ### Skills & Experience: - Demonstrated track record of shipping AI-powered systems to production — we want to hear about systems you have deployed, not just prototyped. - MS, or PhD in Robotics, Computer Science or a related field, or equivalent demonstrated expertise through shipped products. - Deep subject matter expertise in diffusion models applied to robot learning (e.g., diffusion policies, score-based generative models for behavior cloning or planning). - Strong command of optimal control theory and practice, including model predictive control (MPC), trajectory optimization, and feedback control design for physical systems. - Practical understanding of how diffusion-based learning and optimal control approaches are complementary — and the architectural judgment to combine them effectively. - Hands-on experience with hardware integration: sensor pipelines (RGB-D, force/torque, encoders), embedded compute (NVIDIA Jetson, ARM SoCs, FPGAs), and actuator interfaces. - Proficiency in Python and C++; familiarity with ROS 2 or equivalent robotics middleware. - Experience with real-time systems constraints and the performance tradeoffs inherent in deploying learned models on robot hardware. - Strong systems-level thinking — you design for maintainability, observability, and failure modes, not just peak performance. - Excellent communication skills and the ability to drive technical decisions across cross-functional teams. - Willing to work in the office from our Charlestown, MA location at least three days per week.