Field AI is transforming how robots interact with the real world. We are building risk-aware, reliable, and field-ready AI systems that address the most complex challenges in robotics, unlocking the full potential of embodied intelligence. We go beyond typical data-driven approaches or pure transformer-based architectures, and are charting a new course, with already-globally-deployed solutions delivering real-world results and rapidly improving models through real-field applications.
What You’ll Get To Do:
• Design, implement, and optimize 2D/3D CNN and Transformer-based models for deployment on edge and embedded platforms (e.g., NVIDIA Jetson).
• Apply model compression techniques such as quantization, pruning, distillation, and weight sharing to achieve efficient real-time inference under strict constraints on power, bandwidth, and latency.
• Convert, compile, and optimize neural networks for runtime using TensorRT , ONNX , CUDA , and C++ .
• Develop and maintain ROS nodes and interfaces that integrate perception models with the broader robotic system.
• Collaborate closely with AI researchers, robotics engineers, and hardware teams to translate cutting-edge research into deployable solutions on edge devices.
• Build benchmarks, profile and debug runtime issues, and validate performance against real-world scenarios.
• Ensure the reliability, robustness, and stability of deployed models operating in challenging, resource-constrained environments.
The Extras That Set You Apart:
• Familiarity with JAX or additional ML frameworks beyond PyTorch.
• Experience with compiler-level optimizations for GPU inference.
• Background in deploying AI solutions for real-time robotics operating in the field.