01
IRI (CSIC–UPC)
Research project · Diffusion Policies
Working on residual diffusion policies with reinforcement learning for quadruped locomotion in Isaac Lab. The project uses pretrained motion priors and residual control on the SOLO12 robot.
Hi, I’m Sergi Flores.
Robot learning research, computer vision.
AI student at UPC, mostly into robot learning, reinforcement learning and computer vision. I like building things and experimenting as I go, but I’m also interested in research and going beyond quick demos.
Girona, CataloniaBSc Artificial Intelligence · UPC · 2023–2027
$ pwd
/home/sergi/portfolio
$ ls
experience/ education/ projects/ awards/ skills/ about/ cv.pdf
Type help to see the available commands.
01 / experience.log
Experience
02
MANGO
Gen AI Research Intern
Working on diffusion and virtual try-on systems, fashion landmark detection and multimodal computer vision pipelines.
02 / education.md
Education
NTUST · Taiwan Tech
Exchange semester · Erasmus+
UPC · Facultat d’Informàtica de Barcelona
BSc in Artificial Intelligence
03 / projects.json
Projects
01Robotics / Award
Barrendeiros · SENER rover
The complete ROS 2 stack for an autonomous competition rover: RGB-D perception, localization, navigation and sample collection.
02Robotics / Multi-robot
Multi-robot CTF
A real-world multi-robot Capture the Flag system for TurtleBot3 Waffle Pi, combining collaborative SLAM, global and local path planning, computer vision, and autonomous coordination.
03Robotics / Planning
UR3 2×2 Rubik’s Cube solver
A task and motion planning system for solving a physical 2×2 Rubik’s Cube with a UR3 robot, PDDL and OMPL/Kautham planning.
04Robotics / RL
UR3 trajectory tracking with SAC
High-precision trajectory tracking for the Universal Robots UR3 using Soft Actor-Critic (SAC) reinforcement learning in MuJoCo, featuring curriculum learning and reward shaping strategies.
05Machine Learning
Diffusion for CMS collision data
DDPM-based generative model for CMS low-level collision data. Compares diffusion models against GANs and VAEs using statistical metrics (MMD, 1-NN, correlation gap), showing improved fidelity on sparse physics distributions.
06Research / Vision
Equivariant vision for astronomy
Set of experiments on equivariant vision models for image and sequential data using FITS astronomical inputs. Includes training, evaluation, and visualization pipelines for both general and sequential workflows.
073D Vision
Video to semantic 3D
A computer-vision pipeline that turns handheld video into a semantic 3D representation.
08Reinforcement Learning
Reinforcement learning methods
Reinforcement learning repository covering discrete model-free algorithms (Monte Carlo, SARSA, Q-learning) and continuous-control actor-critic methods (DDPG, TD3)
09Hackathon / Vision
NASA Space Apps · AI i Oli
Our team developed a computer vision–powered web application that enables users to explore high-resolution stellar images through natural language queries.
10Hackathon / ML
Medulloblastoma continuum scoring
Semi-supervised continuum scoring for medulloblastoma subtype classification, winner of the BSC challenge at BITS x La Marató 2025.
11Hackathon / Vision
Multi-camera 3D human pose
Real-time human detection, tracking and 3D pose reconstruction from multiple camera streams, awarded second prize at HackEPS 2025.
12Hackathon / ML
CARmen · CUPRA assistant
A multimodal RAG assistant, vehicle-part detector and 3D viewer built for the CUPRA challenge at HackUPC 2025.
13Hackathon / ML
Sempre Visibles
Street-level air-quality forecasting across Catalonia with spatial interpolation, traffic data, Random Forest and LightGBM.
14Hackathon / ML
DataMatch
End-to-end system for participant matching using feature engineering, PCA-based clustering, and cosine similarity. Groups users into balanced teams while considering prior connections and preferences, supported by a web interface and an AI chatbot for explainability and interaction.
Repository data is read from GitHub when the site builds.All repositories ↗
04 / awards.csv
Awards
01 / 07
SENER–CEA Bot Talent · Barrendeiros
Technology + Design
TechnologyDesign
Six months of work integrating perception, localization, navigation and mechanism control into an open-source ROS 2 rover. The team received two of the competition’s three awards: Technology and Design.
view project ↗
BITS x La Marató · BSC Challenge
1st Prize
1st prize
A charity hackathon supporting La Marató de TV3. We won the BSC challenge with five semi-supervised continuum-scoring methods for medulloblastoma subtypes.
view project ↗
HackEPS · INGROUP Challenge
2nd Prize
2nd prize
Second prize for real-time 3D motion capture from two IP cameras, using stereo triangulation and YOLOv8-Pose to create a digital twin of human movement.
view project ↗
NASA Space Apps Challenge
Global Nominee
Global nominee
One of two global nominees at the Barcelona edition. AI i Oli combines classical computer vision with natural-language-to-SQL queries for exploring stellar images.
view project ↗
HackUPC · CUPRA / SEAT Challenge
2nd Prize
2nd prize
Second prize for CARmen, an interactive CUPRA Tavascan manual with a RAG chatbot, visual part search and a 3D interior viewer.
view project ↗
BITS x La Marató · BSC Challenge
1st Prize
1st prize
A charity hackathon supporting La Marató de TV3. We won the BSC challenge Fem visible l’invisible with ML downscaling CALIOPE air-quality data to street level across Catalonia.
view project ↗
Datathon FME · AED Challenge
1st Prize
1st prize
First prize among 300 participants. DataMatch used feature matching, PCA, cosine similarity and a chatbot to find compatible hackathon teammates.
view project ↗05 / skills.yaml
Skills
01
Focus
- Diffusion models
- Reinforcement learning
- Computer vision
- Generative AI
- Multimodal systems
02
Core stack
- Python
- PyTorch
- TensorFlow
- OpenCV
- ROS 2
- Git
- R
- MATLAB
03
Robotics & RL
- Isaac Lab
- MuJoCo
- Gymnasium
- LiDAR
- C++
04
Data & web
- PostgreSQL
- TypeScript
- HTML
- CSS
- Astro
05
Languages
- Catalan · C1
- Spanish · C1
- English · B2
- French · B1
06 / about.txt
Barcelona · 41.38° NAbout
I work across university projects, hackathons, internships and research.
Recently, that has meant diffusion and computer vision work at MANGO, residual diffusion policies for quadruped locomotion at IRI, and the perception stack of an autonomous rover for SENER–CEA Bot Talent. The projects here range from short experiments to work developed over several months.