Projects
Master thesis
Reduced order model abstraction of human nominal locomotion and unexpected down-step occurrences for autonomous bipedal robots. With the developed framework, nominal gaits and compensatory behavior can be embedded on the 3D biped alleviating the need of online trajectory optimization. Using Task-Space Control with ground reaction force embedding, the biped reacts, in a dynamically similar fashion, to the environment that the human was subjected to.
Software: MATLAB, C++, ROS.
Bachelor thesis
Odometry and localization estimation of a driving robot platform. Creation of a multi-phase Linear Kalman Filter framework for ROPOD. Optimality guarantees are maintained by incrementally solving the problem with sensor data that appears linearly in the mapping. Testing and bench-marking of a state-of-the-art infrared ground sensor and comparing it with my algorithm on different surfaces. Algorithm outperformed the $5000 sensor on clean industrial vinyl.
Software: MATLAB, Simulink, Python, ROS.
Autonomous Driving with object recognition
Controlling a simplified Toyota Prius car, by making it able to detect obstacles and people, to avoid objects and stop when people are detected.
OpenCV and PCL libraries are used for human and object recognition in the world
Research Assistant for OpenSim
Developing test cases for speed and accuracy of forward dynamic simulation concerning muscle actuation and wrapping.
Developing an n-dimensional interpolation algorithm for muscle length, lengthening speed, and moment arm evaluation. Speeding up forward dynamic simulations up to three times!
AI Negotiation Agent
An agent has been developed that aims to obtain the best deal for itself. In order to achieve this, strategies for making and accepting bids were developed to respond according to the opponent’s behavior. Since the preferences of the opponent are unknown, it was also imperative to make an approximation based on the information that the opponent gave through bids.
Dynamic Motion Planning for Drones
Master project on quad-copter motion planning and control. Given the structured environment, a rough path is generated using RRT (Rapidly exploring Random Tree). The real-time control is performed by nonlinear MPC (Model Predictive Control) which considers unstructured objects.
5-link bipedal walker
A common bipedal walker is the 5-link walker. In this initial work, I tested different control strategies for generating stable periodic locomotion. Hardware implementations exist such as AMBER-3.
More to come!