SCOPE Lab develops foundational AI methods for decision‑making in large‑scale cyber‑physical systems, applying its research to real‑world challenges in mobility, energy and emergency response. It offers NS‑Gym, an open‑source simulation framework built on OpenAI Gymnasium that isolates environmental dynamics from agent policies, enabling benchmarking of reinforcement learning, planning and continuous‑learning algorithms under non‑stationary conditions. The lab also pilots projects such as AI‑driven multimodal transit integration for U.S. cities.
Funding
Funding not disclosed
Founders
Product
Problem
Decision-making in large-scale cyber‑physical systems such as transit networks, power grids, and emergency response platforms is challenged by constantly changing demand, price fluctuations, and evolving incident patterns. Existing simulation tools often assume stationary environments, making it difficult to develop and evaluate AI agents that can adapt to real‑world non‑stationarity.
Solution
SCOPE Lab creates foundational AI methods that enable adaptive decision‑making in societal‑scale cyber‑physical systems. The lab’s open‑source NS‑Gym framework provides a modular, non‑stationary Markov decision process environment built on OpenAI Gymnasium, allowing researchers to benchmark reinforcement learning, planning, meta‑learning, and continuous‑learning algorithms under realistic change. By separating environment dynamics from agent policies, NS‑Gym supports systematic evaluation of adaptive agents across mobility, energy, and emergency‑response scenarios. The lab also pilots these methods in real communities and organizes tutorials and competitions to accelerate the development of robust, scalable decision‑making solutions.
Target Audience
Primary users are academic and industry researchers developing AI agents for large‑scale cyber‑physical systems, as well as transportation authorities, energy utilities, and emergency‑management agencies seeking adaptive decision‑making tools.
Features
- Open‑source NS‑Gym framework for non‑stationary MDP simulation, compatible with the Gymnasium API
- Modular design that decouples environmental dynamics from agent logic, enabling plug‑and‑play evaluation of RL, planning, meta‑learning, and continuous‑learning approaches
- Built‑in support for realistic scenario changes such as transit demand shifts, energy price volatility, and evolving emergency patterns
- Integrated benchmarking suite with standardized metrics for adaptability, scalability, and performance under non‑stationarity
- Community resources including tutorials (e.g., CPS‑IoT Week) and competitions (e.g., AAMAS 2026) to foster research collaboration and reproducibility
- Proven application in pilot projects for multimodal transit integration, vehicle‑to‑building charging negotiation, and emergency response planning