Projects

• GPU-Accelerated MPC for Neural-ODE Soft Robots

  Continuous-time MPC via Bayesian inference for Neural-ODE-based soft-robot manipulators; achieved 10× accuracy gains and 200× faster solve times vs. CasADi + IPOPT.

• Real-Time Motion Planning for Autonomous Vehicles

  Learning-based vehicle dynamics (GRU, ResNet) for AV trajectory planning; validated on real-world driving datasets. Integrated with MPC benchmarks (CasADi + IPOPT).

  IEEE T-RO 2025 — Paper

• Heavy-Tailed Bayesian Motion Planning (with Student’s-t distribution)

  Sequential Monte Carlo ensemble Kalman-t smoother for robust inference and exploration in deep learning models of vehicle dynamics.

  ACC 2025 — Paper

• GPU MPC of CNNs via Matrix-Var. Ensemble Kalman Smoothing

  Gradient-free CUDA-based matrix-variate EKS for controlling convolutional neural networks; 20× faster and 12× more memory-efficient on large networks.

  ACC 2025 — Paper

• Tensor-Var. GPU EKS for High-Dimensional Deep Learning Models

  Tensor-compatible ensemble Kalman smoother for optimal control of NN-modeled 3D/2D Navier–Stokes & Burgers’ PDEs, enabling millisecond-scale control where classical solvers fail.

• Physics-Constrained Neural-ODE-GRU for HVAC

  Digital-twin HVAC modeling with data-efficient training; 14% accuracy improvement and 5.7× faster runtime.

  MERL Summer 2024

• MERL — Real-Time HVAC Digital Twins (Internship)

  Extended digital-twin concepts to HVAC with real-time Neural ODE/GRU simulators; delivered a PyTorch toolkit and improved forecasting accuracy with GPU acceleration.

  MERL Summer 2024

• MERL — Optimal Inferential Control for HVAC (Collaboration)

  Collaborated with MERL researchers (via KU advisor) on optimal inferential control of NN dynamical systems (HVAC digital twins).

  MERL Spring 2024

• RL-Based Sensor Placement for Persistent Monitoring

  Designed reinforcement-learning policies to minimize estimation-error covariance and improve coverage in monitoring tasks.

• Mobile Robot + SCARA Arm for Ocean Plastic Collection

  Built a mobile robot with SCARA arm; integrated RRT/Dijkstra/potential-field planning with PID control. CAD in SolidWorks, simulation in MATLAB/Simulink.

  Sharif University of Technology 2020–2022

Earlier & Related Work

• Physics-Informed ML for PDEs

  PINNs (FNNs, CNNs, ConvLSTMs) for 2D/3D Navier–Stokes, Burgers, and Kuramoto–Sivashinsky with weak/zero-data training.

• Selected Undergraduate Projects (Sharif, 2018–2021)

  • Ocean plastic collection robot: mobile base + SCARA arm; hybrid planning (RRT, Dijkstra, potential fields) with PID control.
  • Collaborative-robot analysis: kinematics/dynamics modeling and simulation (MATLAB/Simulink); rapid CAD prototyping (SolidWorks).
  • Surgical needle navigation GUI: image-guided toolpath planning with RRT algorithm and visualization interface.