Toshiba Advances Real-Time Autonomous Robot Navigation with Novel Optimization Computer
Summarize this article with:
Toshiba Corporation and MIRISE Technologies Corporation have achieved a breakthrough in autonomous navigation by integrating Toshiba’s Simulated Bifurcation Machine (SBM), a novel quantum-inspired optimization computer, directly into a mobile robot developed by MIRISE. This marks the world’s first successful embedding of such a computer within a mobile platform for autonomous control, overcoming limitations in size, power consumption, and cost that have previously hindered real-time processing of complex information. The companies developed a multi-object tracking algorithm on the SBM, implemented via a proprietary circuit design on an embedded FPGA, allowing the robot to navigate autonomously in dynamic environments. “In multi-object tracking, the association of detected objects with tracked objects are formulated as a combinatorial optimization (matching) problem,” highlighting the core challenge addressed by this innovative system, with results published in Nature Communications and two other academic journals. This advancement promises to accelerate the development of self-driving vehicles and autonomous robots for logistics and smart mobility amid growing demand.
Simulated Bifurcation Machine Embedded in Autonomous Mobile Robotics This advancement addresses the escalating demand for sophisticated processing power within the constraints of size, energy consumption, and cost inherent in self-driving vehicles and robotics, particularly as labor shortages drive adoption in logistics and smart mobility. The core of this innovation lies in a novel multi-object tracking algorithm developed for the SBM, enabling continuous tracking of multiple objects even amidst complex scenarios involving occlusions and route crossings. Current tracking methods frequently falter when objects overlap or become temporarily hidden, but Toshiba and MIRISE’s approach leverages the SBM’s high-speed search capabilities to identify potential “one-to-many matchings,” accurately re-identifying obscured objects. “Re-tracking becomes possible even after something is obscured, improving object motion prediction accuracy,” explains the research team. Evaluations using the Higher Order Tracking Accuracy (HOTA) metric revealed a 4% improvement over standard benchmarks and a significant 23% improvement against newly constructed benchmarks designed specifically for evaluation of object obscuration. Implementation of the SBM on an embedded FPGA allowed for processing at 23 frames per second, exceeding the 10 FPS typically required for automated driving, and demonstrating that “advanced optimization processing—which previously required high-performance servers or dedicated equipment—can be executed in real time on compact, low-power embedded devices.” MIRISE further enhanced path planning by utilizing the SBM’s object tracking data to dynamically adjust object occupancy areas and predict future positions, resulting in more efficient navigation and reduced unnecessary avoidance maneuvers. The companies plan to expand the application of this embedded technology to a wider range of autonomous control scenarios, including cooperative control of multiple robots and route optimization in complex environments. FPGA Implementation Achieves 23 FPS Multi-Object Tracking While existing systems struggle with maintaining continuous object tracking amidst complex scenarios, the companies have successfully embedded Toshiba’s Simulated Bifurcation Machine (SBM) onto an FPGA and integrated it into a MIRISE autonomous mobile robot, achieving a significant leap in real-time performance. Traditional methods often falter in these situations, but the SBM’s capacity for large-scale, high-speed searching allows it to explore potential one-to-many matchings, effectively re-establishing tracking after an object is temporarily hidden. This allows for advanced optimization processing—previously confined to high-performance servers—to occur on compact, low-power devices. MIRISE further enhanced the system with path planning capabilities that utilize object tracking information to dynamically adjust for moving obstacles, resulting in efficient navigation. It analyzes positional confidence and movement direction to dynamically adjust object occupancy areas and predict future positions, reducing unnecessary avoidance maneuvers and securing efficient navigation. HOTA Metric Shows 23% Gain in Obscuration Tracking The collaborative effort addresses a critical challenge in the field: maintaining accurate object tracking in complex, dynamic environments where objects frequently obscure one another. Traditional methods often falter when faced with overlapping or temporarily hidden objects, leading to tracking interruptions and potential safety concerns. Researchers tackled this issue by developing a novel multi-object tracking algorithm leveraging the Simulated Bifurcation Machine (SBM), Toshiba’s proprietary quantum-inspired computer, and implementing it on an embedded FPGA. This innovative approach moves beyond conventional one-to-one object matching, utilizing the SBM’s processing capabilities to explore “potential one-to-many matchings,” ultimately enabling accurate identification even when objects are obscured. This advancement allows for real-time, on-device optimization previously limited to high-performance servers, paving the way for safer and more efficient autonomous navigation in crowded spaces. Unlike quantum computers, they do not require dedicated quantum hardware or specialized peripheral equipment, and operate on standard hardware, such as FPGAs, GPUs, and ASICs. Source: https://www.global.toshiba/ww/technology/corporate/rdc/rd/topics/26/2602-02.html Tags: