my journey in robotics

Co-founder and Project Leader of Driverless UPC.

The first formula-style self driving car developed by students in Spain.

This project consisted on developing a fully autonomous car within just 10 months with a talented team of 20 students.

First we adapted the car with steering and 2 independent brake actuators (due to safety regulations). Then we developed the complete autonomous stack: a Yolo Network to detect the cones delimiting the track, a state estimation algorithm making use of a Kalman Filter. For the global planner we used Voronoi and for the controller we tested with an MPC controller.

During my exchange at ETH, I worked on a framework to autonomously grasp an object (e.g. door handle), but also allowing the operator to take control at any point with a 6 DoF haptic device. One of the key points here was how to transition from autonomous <-> manual mode ensuring the stability of the system. Additionally, we also tackled how to feedback the forces of the robot to the operator, how to develop a User Interface to select the object to grasp in the scene and how to segment and track the object that we wanted to grasp.

During my internship at NASKA.ai (Scaled Robotics) I had the opportunity to develop a ROS-based navigation stack for Jackal to navigate and gather data in construction sites.

We chose a search-based planner for the global path computation and MPC for trajectory optimization and tracking.

This robot navigates in construction sites, where there might  be irregular terrain. In order to ensure smooth trajectories, an elevation map information was added to the planning computation.

As part of a personal project I worked on generating CAD figures from a text input. As opposed to other state-of-the-art methods, this approach does not generate meshes, but actual objects. It uses a self-trained AI model to decode the primitive shape and parameters from a text input. Then, it generates a 3D CAD file, containing the shape specified as text.

The idea behind it is to speed up the design and allow a smooth introduction into CAD design to people who are not use to it.

The source code is available here: https://github.com/monicapserrano/text_to_cad

Probably the most unexpected robot you will find in the internet today. This (experimental) bio-inspired design consists on several modules that can work independently or attached to one another. The default configuration can run on different terrains. The upper shell can rotate up to 180°, allowing the robot to navigate in water, as well.  

Why a Mantis Shrimp? After asking ChatGPT which bio-inspired robot has not yet been designed, I got this crustacean as a response.

During my years in robotics, I have developed and implemented deterministic (e.g. Pure Pursuit Controller) and non-deterministic (e.g. MPC) controllers for 2D navigation.

My designs focus on achieving maximum smoothness, without compromising tracking accuracy.

Ensuring smoothness not only portrays the robot as a safe machine, but it also allows one same controller to be used in very different robots with minimum to no tuning.

As part of my Bachelor Thesis in Mechanical Engineering, I designed the layout of the sandwich structure (carbon fiber + aluminum honeycomb) for a Formula Student car chassis. The software used was Hypermesh, a Finite Element Analysis software. Some of the biggest challenges here were how to simulate the forces the car is withstanding while driving at high speeds and how to accurately produce the monocoque (which was manually manufactured by us).