CV
Education
- Ph.D., Mechanical and Aerospace Engineering, Arizona State University, Aug 2025 – present
- Advisor: Prof. Kunal Garg — GPA 4.25/4.33
- M.Tech., Electrical Engineering, Indian Institute of Technology Gandhinagar, 2018 – 2020
- Advisor: Prof. Naran Pindoriya — GPA 9.43/10
- B.E., Electrical Engineering, Gujarat Technological University, 2014 – 2018
Experience
- Graduate Research Assistant, Safe Robotics Lab (SRG), Arizona State University, Aug 2025 – present
- Dissertation: learning-based predictive safety for autonomous systems — enforcing barrier constraints over predicted horizons and diagnosing failures in learned controllers, scaling from single-agent control to distributed multi-robot teams under communication delay.
- Researcher, Data and Decision Sciences, TCS Research, Mumbai, Sept 2020 – July 2025
- Mentors: Dr. Mayank Baranwal, Dr. Harshad Khadilkar.
- RL for safety-critical networked infrastructure under uncertainty and adversarial attack: multi-robot coordination, power grids, railway scheduling, and supply chain. NASSCOM AI Gamechangers Award, top 10 AI projects in India (2024).
- Adjunct Faculty, B.Sc. Data Science, S.P. Jain School of Global Management, Mumbai, Dec 2023 – July 2025
- Taught Introduction to Data Science and Introduction to Programming; supervised undergraduate projects.
Selected projects
- BarrierFormer — transformer-guided predictive barrier enforcement (CoRL 2026, main track)
- Encoded horizon-level discrete-time CBF constraints into policy parameters using an SQP-based safety teacher and a barrier critic over transformer-predicted rollouts, removing online optimization and model knowledge at inference.
- Raised safety rate to 96% across 2D/3D and linear/nonlinear dynamics against 65–76% for diffusion- and transformer-based baselines; matched model-based MPC on safety while cutting per-step latency by three orders of magnitude (5 µs vs 8.3 ms); generalized zero-shot to larger, denser workspaces.
- FIND-R — failure identification and directed repair (under review)
- Localized safety-encoding failures in GNN controllers (GCBF+, InforMARL) by probing layer-wise representations, tracing the bottleneck to an unsupervised attention mechanism and repairing it with a supervised radial-attention head — cutting false modes by 95% and 65% respectively.
- Ran end-to-end in under 9 s where SMT and CROWN formal-verification baselines exceeded a 4-hour budget.
- MRTAgent — safe multi-robot task allocation (ECAI 2024; AAMAS 2025; patents)
- Dual-agent RL framework inspired by self-play for real-time fleet task allocation, coupled with a modified LQR controller for collision-free navigation; safety enforced at the control layer while the learned policy handled assignment.
- PowRL — constraint-safe RL for power grids under adversarial attack (AAAI 2023; patents)
- Heuristic-guided safety filter masking the action space to a verified-feasible subset, cutting 72,000 actions to 240 on the IEEE 118-bus network without violating operational limits.
- Ranked 1st on L2RPN NeurIPS 2020 (Robustness) and 3rd of 90 teams in L2RPN 2023 (TU Delft / RTE-France).
Publications
Full list on the publications page.
- BarrierFormer: Transformer-Guided Predictive Barrier Enforcement for Safe Robot Control — Conference on Robot Learning (CoRL), main track, 2026
- PRIORITY-Q: Certifiable Event-Queue Scheduling for Real-Time Railway Resource Allocation — International Conference on Neural Information Processing (ICONIP), main track, 2026
- Together We Rise: Optimizing Real-Time Multi-Robot Task Allocation using Coordinated Heterogeneous Plays — AAMAS, AAAI Track, 2025
- Optimizing Multi-Robot Task Allocation in Dynamic Environments via Heuristic-Guided Reinforcement Learning — European Conference on Artificial Intelligence (ECAI), 23% acceptance, 2024
- Peer-to-Peer Energy Trading Framework: An Experimental Evaluation — 13th IEEE PES ISGT Asia, 2024
- A Learning Approach for Cost-Efficient Sourcing and Routing Strategies in E-Commerce — CODS-COMAD, 2024
- Multi-Agent Learning of Efficient Fulfillment and Routing Strategies in E-Commerce — NeurIPS Workshop on Generalization in Planning, 2023
- PowRL: A Reinforcement Learning Framework for Robust Management of Power Networks — AAAI Conference on Artificial Intelligence, 19.2% acceptance, 2023
- Real-time Simulation of V2G Operation for EV Battery — 21st National Power Systems Conference (NPSC), 2020
Patents
Five filings across three inventions (three India, two U.S. applications) — see the patents page.
Awards
- NASSCOM AI Gamechangers Award, 2024 — top 10 AI research projects in India
- Third Prize, L2RPN Challenge 2023 — 3rd of 90 teams (TU Delft / RTE-France)
- POSOCO Power Systems Award, 2021 — one of 15 recipients in India for an M.Tech thesis in power systems
Technical skills
- Languages: Python, MATLAB, SQL
- Learning: PyTorch, JAX/Flax, TensorFlow, Gymnasium
- Optimization: CasADi, IPOPT, sequential quadratic programming, Z3, dReal
- Tools: Git, Linux, AWS, LaTeX