Portfolio

Apples classifier

During my bachelor in mechatronics I participated in a Computer Vision Project to classify apples trough a Neural Network using OpenCV. Document

Segmentation
Apple segmentation ROI
NN
Neural Network classification
Dataset
Apples labeling

Delta Robot

My bachelor's thesis consisted of a Delta Robot detailed design. The robots dimesions were obtained using a basic optimization algorithm that maximized the robots workspace. The mechanical loads were goten using NX Nastran, which allowed the motor+controller selection process. Document

TerstFcn
Delta robot final design
Evol
Delta robot workspace
Sim
Delta robot components

3D Laser Scanner

I worked as a R&D engineer for almost 3 years in an industrial automation company (Colsein). I participated in the design of a 3D laser scanner for tank calibration, I did mechanical design, manufacturing, electronic integration, point cloud and mesh analysis, software and calibration reports. The scanner was partially funded by the Colombian Institute for Science "Colciencias".

RS
Tank gauging process
RS2
Tank point cloud representation
Real
Final laser scanner

Universidad Nacional Assitant Teacher

I came back to academia as an assitant teacher in Robotics, I worked for 2 years, as a team with Prof. Pedro Cardenas, we changed the class, we introduced new tools as ROS, Robotics Toolbox fro Matlab, and we managed to promote new ideas in the students. I loved the job,but it lasted 2 years (from 2017-2 to 2019-1), it was probably one of the most beloved times in my life.

Classes repos: Github
Hans
2017-2 class
2018-2
2018-1 class
Best group
2018-2 class

PSO applied on path planning

In my Master's degree I was involved with path planning for the industrial robot ABB IRB 140. I implemented a Particle Swarm Optimization algorithm to create colision free paths for basic motion tasks. Te code was developed in Matlab and the connection to the robot used a postprocessed code programmed in RAPID (Programming Language for ABB Robots). Document

Test Fnc
Test function for the PSO
Particles
Particles path evolution
Layout
PSO generated path
Sim
Simulated path generated by the postprocessed RAPID code
Sim2
Simulated path generated by the postprocessed RAPID code
Real
Real robot executing computed path

Motion planners benchmark

I am really into path planning for robots, actually it has been the main topic in both, my bachelor's and my master's thesis. I got involved with modern robotics tools such as ROS, Gazebo, Moveit, among others. In association with Sebastian Realpe, we have developed some packages for the ABB Robots located in the LabSIR. But the main job wwas the adaptation of a powerful tool called robowflex to a set of task for motion planning benchmarking.

ABB Packages repo: Github
Benchmark repo: Github
Plot Box
Boxplot: Planners time performace
Waypoints
Radarplot: Planners waypoints
Runs
Radarplot: Planners success rate
Time
Radarplot: Planners computation time

Frobs_RL - Reinforcement Learning library for robotics

It is a project developed mainly by Jose Manuel Fajardo as an spinf-off project that we proposed. FRrobs_RL is a library intended to connect Reinforment Learning Algorithms (TD3, PPO, SAC) to robotics enviroments in Gazebo. It allow to deploy simulated models to real hardware using ROS.

FRobs_RL repo: Github
FRobs_RL resources repo: Github
Training
ABB IRB120 Training process

Affordances study for informed motion planners

It was my main research topic for the PhD. It integrates knowledge about an industrial robot enviroment to compute some affordances of thej obtects to manipulate and then incorporate inforamtion to the motion planner to execute an adeccuate task.

Vision A kinvect v.1 was used to adquire a rgb and deep images from the enviroment. A NN based on perceptron was used to segment the regions of interest. Link to agnostic segmentation.
Grasping The Grasping Generation Convolutaional Neural Network was used to propse multiple grasping hypothesys. Link to GGCNN.
Motion Planning: A modified probabilistic motion planner from moveit was selected.