GTC 2020: Terrain Traversability Estimation using Normal Distribution Transform on GPU in Amazon Scout
After clicking “Watch Now” you will be prompted to login or join.
Click “Watch Now” to login or join the NVIDIA Developer Program.
Terrain Traversability Estimation using Normal Distribution Transform on GPU in Amazon Scout
Ka Chen, Amazon
Scout is a autonomous robot from Amazon for delivering packages. Mapping its surroundings in real time to identify traversable space is critical for autonomous driving. We'll discuss how we implemented a technique called "3D Normal Distribution Transform" (3D NDT) on NVIDIA TX2 to calculate surface traversability and help Scout navigate the neighborhood autonomously and safely. First, we'll introduce what 3D NDT is and what it is used for. Then we'll discuss how we implemented NDT in Amazon Scout to calculate surface traversability to help the robot to avoid obstacles and navigate safely. We'll highlight how we implemented NDT on NVIDIA TX2 GPU, achieving more than 10x faster run time than before, and how that greatly improved Scout's navigation. We'll also go into the challenges and innovation involved.