Developed a multi-agent framework using an autoregressive token-based motion predictor with Priority-Based Search to coordinate safe autonomous vehicles amidst human drivers, reducing collision rates and travel times.
Developed a physics-informed meta-learning framework with conditional, variational, and flow matching-based Neural Process variants to eliminate the retraining bottleneck of standard PINNs, enabling few-shot adaptation.
Implemented DreamerV2 world model architecture with a self-supervised DINO backbone to evaluate latent representations from 3D first-person inputs within the ViZDoom environment for benchmarking.
Developed an actor-critic agent (PPO) applied on a mobile robot for exploration and navigation tasks. Intrinsic rewards formulated from Random Network Distillation module. Tech: OpenAI Gym, PyTorch, ROS
Using Online Sequential Extreme Machine Learning for real-time uncertainty correction in the Adaptive Cruise Control problem. Project part of the Robotics Program RISS at Carnegie Mellon University. Tech: Torch
A PD with Gravity Compensation and Sliding Mode Control Comparison. Trajectory tracking comparison between PI and SMC under gaussian disturbances. ICCAD 2020. Tech: ROS
Transforming torque inputs into position inputs for a manipulator robot in cleaning window task. Proposed hybrid controller composed of a learning approach and inverse dynamics. CDC 2020.
Robot platform based on haptic technology for surgery tasks based on UR5 manipulator robots. Project fully funded by CONCYTEC.