07-30, 13:30–14:00 (UTC), Purple Track
In robotics, the two language problem is potentially even more pronounced than in other fields of scientific computing. KEF Robotics is using Julia to tackle this challenge and rapidly develop, field test, and commercialize autonomy software for small multirotor drones.
There is currently a very large gap from new robotics research to commercial products. The two language problem contributes to this gap, where research and prototyping often occurs in MATLAB or Python and then field testing and commercialization requires porting to C or C++. The development speed, runtime speed, and scientific computing capabilities of Julia offer a promising solution to this challenge.
KEF Robotics is developing an attachable autonomy subsystem for small aerial vehicles, allowing us to add navigation, hazard avoidance, and machine learning capabilities to any drone, all with just cameras (no GPS required). Our hazard detection and avoidance module has been developed, tested, fielded, and is now being transitioned into a commercial product, all in Julia. This module combines widely varying components, including image processing, geometric computer vision, splines, and nonlinear optimization, making it a perfect showcase for Julia's unique strength in composability and scientific computing.
Another key challenge in robotics is predictable real-time performance, often on embedded computing. These factors are typically not associated with garbage-collected scripting languages. While Julia's tooling and runtime are not perfect for the task, options for allocation management, profiling, and package compilation allow prototype code to be ported to these applications, all in the same language. Also, Julia native CUDA programming is of particular utility for NVIDIA's Jetson products, which are prevalent in robotics research and commercialization.
Co-founder at KEF Robotics. Formerly R&D at Astrobotic. B.S. in Computer Science and Robotics at Carnegie Mellon University.