Senior Applied Scientist, Last Mile Robotics

Every doorstep is different. Our robot has to understand all of them. We're building robots to deliver packages to customer homes. To make that possible, we're creating a system that can automatically generate a simulation-ready digital twin of the outdoor delivery environment for every customer home Amazon delivers to; modeling the driveways, yards, walkways, stairs, porches, and doorsteps a robot must understand and navigate. And the goal isn't simply to create virtual worlds that look real. They need to be real in the ways that matter to a robot. Geometry, semantics, appearance, physical properties, and the varied environmental cues that influence perception, planning, and motion, must be reproduced faithfully so robot behaviors from the field are repeatable in simulation. Our simulator becomes an effective tool to study the problem. The same autonomy software that runs on the physical robot also runs in simulation. When robot behavior differs between simulation and reality, we can ask why, and improve the representation, simulation, or models that produced the difference. Our North Star ambition is a system capable of modeling the delivery environment for every customer home we serve, across the enormous diversity of real residential environments. There isn't an established recipe for doing this. We are advancing the knowledge frontier of scene reconstruction. The right representation may combine classical geometry, learned scene representations, generative models, world models, or approaches that haven't been invented yet. We're interested in people with deep expertise in one or more areas of: Computer vision, 3D graphics, scene reconstruction, physics simulation, neural scene representations, Gaussian splatting, generative AI, world models, or related areas - with a strong interest in applying that expertise to robotics.

Overview This is applied research with a short distance from idea conception to physical consequence. You'll work closely with robotics, autonomy, simulation, and software engineers to turn promising methods into systems that change how real robots perceive, plan, and navigate at Amazon scale. As an Applied Scientist on the Last Mile Robotics team, you will develop cutting-edge methods for capturing, reconstructing, and simulating real-world delivery environments in high-fidelity 3D. Your work will directly enable autonomous robots to understand and navigate the diverse conditions found at customer doorsteps. You will push the boundaries of scene reconstruction, neural rendering, and generative modeling, transforming scientific breakthroughs into physical impact through rapid experimentation and close collaboration with robotics, autonomy, and simulation teams.

We're a multidisciplinary team working across robotics, simulation, computer vision, graphics, machine learning, and software engineering. We don't expect the best ideas to come from any one discipline. A graphics researcher, roboticist, computer-vision scientist, simulation expert, or machine-learning researcher may see a different path through the same problem, and we expect the strongest solutions to emerge from bringing those perspectives together. We value technical depth, curiosity, rapid experimentation, and low-ego collaboration. We ask for help when we need it and give it generously when someone else does. We like problems where the answer isn't obvious and turning promising ideas into working systems. Every doorstep is different. Help us build the technology that lets a robot understand all of them.