Compositional Policy Learning in Stochastic Control Systems with Formal Guarantees
Žikelić, Đorđe, Lechner, Mathias, Verma, Abhinav, Chatterjee, Krishnendu, Henzinger, Thomas A.
–arXiv.org Artificial Intelligence
Reinforcement learning has shown promising results in learning neural network policies for complicated control tasks. However, the lack of formal guarantees about the behavior of such policies remains an impediment to their deployment. We propose a novel method for learning a composition of neural network policies in stochastic environments, along with a formal certificate which guarantees that a specification over the policy's behavior is satisfied with the desired probability. Unlike prior work on verifiable RL, our approach leverages the compositional nature of logical specifications provided in SpectRL, to learn over graphs of probabilistic reach-avoid specifications. The formal guarantees are provided by learning neural network policies together with reach-avoid supermartingales (RASM) for the graph's sub-tasks and then composing them into a global policy. We also derive a tighter lower bound compared to previous work on the probability of reach-avoidance implied by a RASM, which is required to find a compositional policy with an acceptable probabilistic threshold for complex tasks with multiple edge policies. We implement a prototype of our approach and evaluate it on a Stochastic Nine Rooms environment.
arXiv.org Artificial Intelligence
Dec-3-2023
- Country:
- Oceania > Australia
- New South Wales > Sydney (0.04)
- North America
- The Bahamas (0.04)
- United States
- Maryland > Baltimore (0.04)
- Tennessee (0.04)
- District of Columbia > Washington (0.04)
- Hawaii > Honolulu County
- Honolulu (0.04)
- Rhode Island > Providence County
- Providence (0.04)
- Louisiana > Orleans Parish
- New Orleans (0.04)
- Pennsylvania > Centre County
- University Park (0.04)
- Massachusetts > Middlesex County
- Cambridge (0.14)
- California > Los Angeles County
- Los Angeles (0.28)
- Long Beach (0.04)
- Mexico > Quintana Roo
- Cancún (0.04)
- Canada
- Quebec > Montreal (0.04)
- British Columbia > Metro Vancouver Regional District
- Vancouver (0.04)
- Alberta > Census Division No. 15
- Improvement District No. 9 > Banff (0.04)
- Europe
- Austria > Vienna (0.14)
- Germany > Berlin (0.04)
- Czechia > Prague (0.04)
- United Kingdom > England
- Greater London > London (0.04)
- Cambridgeshire > Cambridge (0.04)
- Switzerland > Zürich
- Zürich (0.14)
- Russia > Northwestern Federal District
- Leningrad Oblast > Saint Petersburg (0.04)
- France
- Île-de-France > Paris
- Paris (0.04)
- Provence-Alpes-Côte d'Azur > Alpes-Maritimes
- Nice (0.04)
- Île-de-France > Paris
- Asia
- Singapore (0.04)
- Russia (0.04)
- Myanmar > Tanintharyi Region
- Dawei (0.04)
- Middle East > Israel
- Haifa District > Haifa (0.04)
- China > Shaanxi Province
- Xi'an (0.04)
- Oceania > Australia
- Genre:
- Research Report (0.70)
- Technology: