Customizing chips for a highly specific purpose is not new, but as companies and technology experts look to perform more local computation on devices with more conservative power and computing constraints, rather than sending data back and forth to large data centers via network connections, the approach involves creating hyperspecific chips designed based on the robot's physical layout and its intended use. By considering the robot's perception of its surroundings, its mapping and understanding of its position within those environments, and the requirements for motion planning resulting from these mappings and desired actions, researchers can design processing chips that significantly improve the efficiency of this final stage by supplementing software algorithms with hardware acceleration.
The classic example of hardware acceleration that most people encounter frequently is the graphics processing unit, or GPU. A GPU is essentially a processor specifically designed to handle graphics computation tasks, such as display rendering and video playback. GPUs are popular because almost all modern computers run graphics-intensive applications, but custom chips for a range of different functions have recently become more popular due to the emergence of more customizable and efficient small-chip manufacturing technologies.
MIT News has now introduced the design of the Neuman robot control hardware chip. This system creates a custom hardware design best suited to meet the computational needs of a specific robot. Users input parameters for the robot, such as its limb layout and the movement of its joints. The Neuman system transforms these physical properties into mathematical matrices. These matrices are "sparse," meaning they contain many zero values that roughly correspond to movements impossible under a given robot's specific anatomy.
The system then designed a specialized hardware architecture that only performs calculations on non-zero values in the matrix. Therefore, the resulting chip design was tailor-made to maximize efficiency for the robot's computational needs. This customized design paid off in testing. Neuman's team used a field-programmable gate array (FPGA), which is somewhat of a middle ground between a fully custom chip and an off-the-shelf CPU, and its performance was significantly better than the latter.
Enabling robots to react more quickly to their environment is not just about increasing manufacturing speed and efficiency, but also about making robots safer to work alongside and collaborate with humans. This remains a significant obstacle to the wider adoption of robotics in everyday life, meaning this research could help ensure harmonious coexistence between humans and robots in the future.
