On June 19th, 2023, Brian Plancher, assistant professor of computer science, and colleagues presented new research, titled “RoboShape: Using Topology Patterns to Scalably and Flexibly Deploy Accelerators Across Robots,” at the International Symposium on Computer Architecture (ISCA) in Orlando, Florida. The work the team presented at this year’s conference advances the field of computer architecture for robotics through a generalizable and automated accelerator framework.
Specifically, RoboShape enables engineers to directly use robot topology to inform architectural mechanisms including task scheduling and allocation, data placement, block matrix operations, and sparse I/O data. To accomplish this, the researchers leveraged two topology-based computational patterns that scale with robot size: topology traversals and large topology-based matrices. Professor Plancher and his colleagues assert that one of the key challenges for hardware acceleration of robotics applications is the enormous diversity of possible robot designs and deployment scenarios. As a generalizable and automated accelerator framework, RoboShape helps address this problem by making it possible to effortlessly embed domain-specific insights directly into the accelerator design process for robotic systems.