New automata-based approach enhances reinforcement learning for complex control systems

Researchers have developed a new automata-based approach for control synthesis using Signal Temporal Logic (STL). This method addresses challenges in reinforcement learning (RL) for complex systems lacking accurate models by providing an efficient memory mechanism and associated Markovian rewards. The approach constructs a timed alternating automaton from STL specifications, augmenting the state space with automaton locations and clock valuations to derive rewards from the acceptance condition. Empirical results show this method outperforms existing approaches in learning policies with higher robustness scores and satisfaction rates. AI

IMPACT This research could enable more robust and efficient control policies for complex AI systems, particularly in scenarios with incomplete system models.

RANK_REASON This is a research paper detailing a novel technical approach in AI. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Alper Kamil Bozkurt
• arXiv
• Markovian rewards
• reinforcement learning
• Signal Temporal Logic
• timed alternating automaton

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →