Overall, I enjoy being a full stack roboticist, and I have had the opportunity to work on a variety of robot systems. My favorite achievement is developing a mobile manipulator smart robotic assistant to collaborate with humans in industrial environments. This involved working with a UR5 manipulator, end-effectors, a custom-built holonomic mobile base, and sensor systems. I had to implement a SLAM system and ROS navigation stack on the holonomic mobile base, which enabled mapping, localization, and path planning in a warehouse domain. I also implemented a ROS MoveIt motion planning pipeline for both pick and place operations and scanning operations. Lastly, I have integrated a task and motion planning pipeline that generates a task and motion plan given a set of tasks and the human's needs.

We are collaborating with Dr. Stefanos Nikolaidis and the ICAROS lab in developing algorithms that can generate contingency-aware task plans, predict human actions and adapt robot actions to ensure safe and efficient human-robot collaboration in assembly manufacturing. We do this by learning the human operator's individualized preferences and enabling the robot to adapt to its collaborator, anticipate actions and take supportive actions.  We show a video demonstrating this system.

We present a framework for human-robot collaboration in high-mix applications where tasks requiring delicate or complex operations are assigned to humans, while straightforward tasks are assigned to robots. The framework includes the use of simulation to design the assembly cell, computer vision for part localization, and automated motion planning to generate robot trajectories. The cell is also capable of tracking humans for safety purposes, and features a unique contingency detection system to alert operators in case of unexpected situations. A satellite assembly case study is used to demonstrate the capabilities of the cell and the feasibility of performing complex assembly operations using human-robot collaboration. 

In this work, we focus on the use of robots for external heating of components in manufacturing. To efficiently position the heating tool, we present a physics-aware action selection policy that employs forward simulation with a branch and bound search to determine the best action sequence. We also propose a parameter learning graph-based modeling framework that utilizes physics-informed machine learning to predict the temperature evolution of the surface with respect to time. Furthermore, we introduce a state transition model to describe how the thermal characteristics of the system change based on the heating tool's position. Our proposed approach is demonstrated on an industrial tool for the composite layup process, and we show the usefulness of physics-inspired machine learning in a real-world application.