Reproduce “Dynamic-Weighted Simplex Strategy for Learning Enabled Cyber Physical Systems” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shreyas Ramakrishna, Charles Hartsell, Matthew P Burruss, Gabor Karsai, Abhishek Dubey
A dynamic-weighted simplex strategy, developed by researchers at Vanderbilt University's ISIS, leverages reinforcement learning to smoothly blend control outputs from a learning-enabled component and a baseline controller. This method reduced out-of-track occurrences to 3 events compared to 8-10 for standalone controllers, enabling higher average operating speeds on an autonomous car testbed, with resource management handling computational demands.
Key insights
- Dynamically blending controller outputs via RL can achieve a superior balance between performance and safety in LEC-enabled CPS compared to discrete switching or fixed weighting.
- Reinforcement Learning provides an adaptive mechanism for decision logic in safety architectures, effectively learning optimal trade-offs in complex, dynamic environments.
- Practical deployment of sophisticated safety architectures on resource-constrained hardware is feasible through integrated resource management, including predictive thermal management and task offloading.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.