Robots: A Growing Force in Our World
Robots are becoming increasingly integral in various sectors such as economy, healthcare, agriculture, and defense. With predictions of over 20 million robots in operation by 2030 and a burgeoning robotics market, their seamless integration into real-world settings is more crucial than ever. This integration requires robust autonomous capabilities and the ability to execute complex AI tasks in real-time.
The Surprising Gap in Research
Despite the escalating importance of robotics, there’s a noticeable disconnect between the robotics and computer architecture research communities. Among the myriad papers published in top architecture conferences in 2023, only a few addressed robotics. This gap stems in large part from the absence of a comprehensive benchmark suite that could clarify the performance needs of robotic tasks and their architectural implications.
Limitations of Existing Benchmark Suites
Current benchmark suites, while valuable, often miss crucial aspects necessary for architecture-robotics research. Narrow scopes, programming language limitations, and a lack of modern software techniques are some of the shortcomings.
Introducing the Robotics in the Wild (RoWild) Benchmark Suite
RoWild fills this gap by providing a broad range of robotic tasks, covering the complete software pipeline of autonomous robots: perception, planning, and control. Continuously, sensory data is read and processed through this pipeline, generating commands for the robot’s actions. RoWild’s design ensures versatility, enabling users to configure and sequence tasks to model various robotic applications.
This suite is derived from an analysis of 29 industrial robots, ensuring its relevance and breadth in the selection of tasks and algorithms. It also integrates research proposals, such as deep learning-based planning, to represent the state-of-the-art in research.
RoWild’s Features
- Comprehensive: RoWild covers a wide range of robotic workloads, which is vital for the development of versatile hardware solutions.
- High-Performance: Designed with performance optimization in mind, RoWild uses native languages and industry-standard profilers to enhance efficiency.
- Simulator-Friendly: Compatibility with architectural simulators makes RoWild a valuable tool for hardware research. RoWild is by default integrated with ZSim.
- Cross-Platform: RoWild caters to a wide range of compute resources by developing applications in C++, CUDA, and Verilog, reflecting the heterogeneous nature of modern AI/robotic platforms.
- Versatile and Modular: The suite’s versatility and modularity enable researchers to “mix and match” tasks, offering a realistic approach to modeling end-to-end robotic applications. The RoWild repository includes six examples of such comprehensive robotic applications.
Accessing RoWild and Further Reading
The RoWild suite is available for both access and contributions on GitHub. For a comprehensive overview of our methodologies and results, please refer to our research paper, which has been published in the Proceedings of the ACM on Measurement and Analysis of Computing Systems (POMACS). Additionally, RoWild will be featured at the SIGMETRICS 2024 conference, scheduled for June 2024.
About the Authors
Mohammad Bakhshalipour, a fifth-year PhD student, and his advisor, Professor Phillip B. Gibbons, lead this initiative at Carnegie Mellon University. Their research bridges the gap between robotics and computer architecture.
This blog post is crafted to offer a succinct yet thorough overview of RoWild, aiming to engage the broader computing community and spark interest in the intersection of robotics and computer architecture.
Disclaimer: These posts are written by individual contributors to share their thoughts on the Computer Architecture Today blog for the benefit of the community. Any views or opinions represented in this blog are personal, belong solely to the blog author and do not represent those of ACM SIGARCH or its parent organization, ACM.
