Navid Azizan Ruhi
Massachusetts Institute of Technology, MA·Engineering·Early-career
Self-driving vehicles, robots and power-grid controllers increasingly learn from data. Yet the machine-learning parts inside them come with few safety guarantees.
- Fields
Engineering
Computer Science
Mathematics & Statistics
- Topics
safe reinforcement learning
autonomous systems
trustworthy AI
- Grants
- 1 known grant, through Mar 2031
What they work on
- Develop Hard-Constrained Neural Networks (HardNet), which use a differentiable projection layer to guarantee input-dependent constraints, and embed them in reinforcement learning to produce certifiably safe policies.
- Use HardNet for boundary control of systems governed by partial differential equations, enforcing Lyapunov stability conditions as hard constraints so the controlled system is guaranteed to settle.
- Create run-time uncertainty measures for pre-trained perception models, using neighborhood-consistency tests and curvature sketches, to detect out-of-distribution inputs without retraining and to guide robust adaptive safe reinforcement learning.
- Scale safety to many agents through decentralized multi-agent reinforcement learning with Hamilton–Jacobi reachability, federated RL for agents with different dynamics, and online incentive mechanisms, validated in robotics, transportation and building-energy testbeds.
More about the research
They may break safety rules, fail when conditions differ from their training data, or struggle when many agents must coordinate. In public spaces and critical services, even rare mistakes can injure people or set off cascading failures. This project aims to build the math and algorithms that make safety a built-in feature of learning systems. That ranges from single neural networks, to learned controllers, to large systems with many agents.
Why it matters
As learning-based autonomy enters roads, robots and building systems, failures can harm people or disrupt critical services. Methods that guarantee constraints are met, detect unfamiliar situations, and coordinate many agents safely could make autonomous transportation and robotics safer and building energy use more reliable and efficient. The project also plans open-source tools and datasets.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Vijay Subramanian
University of Michigan, MI·Engineering
Develop robust reinforcement learning algorithms that use model-class knowledge, future predictions and off-policy methods to cope with mismatches and limited data from simulators or offline records.
Ioannis Kantaros
Washington University in St Louis, MO·Engineering·Early-career
Train multi-agent reinforcement learning (MARL) control policies for robot teams and mathematically certify their robustness against adversarial, high-dimensional sensor inputs, testing them on aerial and ground robot testbeds and photorealistic simulators.
Yuxin Chen
University of Pennsylvania, PA·Mathematics & Statistics
Characterize the fundamental limits on sample and computational complexity of reinforcement learning, meaning how much data and computing an agent needs to learn good decisions in an unknown environment.
The school
Massachusetts Institute of Technology: 25 labs on PI Pages, $1.1B a year on research, 692 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.