ROMANUS: Dynamic Neural Architectures for Autonomous Systems
Tech ID: 34078 / UC Case 2023-713-0
Brief Description
ROMANUS is a cutting-edge methodology designed to enhance the performance and robustness of latency-critical,
real-time intelligent systems through dynamic neural architectures.
Full Description
ROMANUS introduces a novel approach for designing and deploying multi-sensor autonomous systems, such as
autonomous vehicles (AVs) and unmanned aerial vehicles (UAVs), with an emphasis on dynamic neural network
architectures. It uniquely adapts to various operational modes to optimize efficiency and robustness, outperforming
existing methods in autonomous systems by improving performance, energy efficiency, and prediction quality.
Suggested uses
- Design and development of autonomous vehicles by companies like Waymo, Argo AI, and Tesla.
- Enhancements in AR/VR systems for tech giants such as Facebook Reality Labs, Google, and Apple.
- Deployment in critical applications by the U.S. Department of Transportation and the U.S. Department of Defense.
- General technology advancements in AI and autonomous systems across various sectors.
Advantages
- Superior performance, energy efficiency, and robustness in autonomous systems.
- Innovative multi-branch design for optimizing multi-sensor system operations.
- Adaptive operational modes for diverse contexts and deployment conditions.
- Efficient task offloading by understanding and adapting to the deployment environment.
- Lightweight capture of spatiotemporal correlations to adapt operational modes and execution branches.
- Low-overhead monitoring of deployment conditions and processing branch performance.
Patent Status
Patent Pending
Related Materials
- Odema, M., Chen, L., et al. Al Faruque, M. (2022). Testudo: Collaborative Intelligence for Latency-Critical Autonomous
Systems. IEEE TCAD. 42 (6).
- Chen, L., Odema, M., et al. Al Faruque, M. (2022). Romanus: Robust Task Offloading in Modular Multi-Sensor
Autonomous Driving Systems. 2022 IEEE/ACM ICCAD.