Andreu Huguet Segarra

andreu.huguet@estudiantat.upc.edu Huguet57 andreuhuguet

Projects

LIMO-Velo: LiDAR-Inertial SLAM for fast and aggressive movement

September 2021 - March 2022 @ HKUST as a Visiting Intern Researcher

Based on Fast-LIO's approach, this framework corrects the localization with LiDAR observations at an extremely high-frequency (up to 200Hz) and maps directly to an Incremental KD-Tree at low-frequency (10Hz). Allowing us to post-process the pointcloud accordingly (remove dynamic objects or apply segmentation) while not losing any quality w.r.t. to mapping at 200Hz.

We call it “Localize Intensively and Map Offline” (L.I.M.O.) and instead of waiting for the full pointcloud, it uses the points as a continuous stream. This allows us to go at high velocities (20m/s) and aggressive turns (120deg/s) with a centimeter-level accurate dense map.

I was advised by the most talented research mentor Qifeng Chen throughout the whole process and I published the code online for anyone to use and improve on: https://github.com/Huguet57/LIMO-Velo/.

Main contributions

+ First algorithm using the fact that spinning LiDARs can localize at higher frequency than mapping without loss in quality.
+ With higher localization frequency, the steps are smaller, less complex and better approximated by EKF (since it assumes quasi-linearity).
+ More accurate motion compensation that can be smoothed out and processed when mapping.
+ Ability to post-process what it maps while being real-time.

XALOC: Spain’s first ever driverless vehicle

September 2020 - September 2021 @ UPC as a Student

XALOC’s journey began in 2018 when within a full year, extremely talented engineers built the first DV vehicle ever seen in Spain and finished 10 laps in an official Formula Student Driverless competition (in 20 minutes).

After the pandemy hit Spain the hardest, the project had to be stopped in March 2020 and we resumed it in September 2020 with a clear goal in mind: reach 90km/h with a reliable autonomous solution.

In May 2021, the car reached its highest velocities of 60km/h before disaster strucked: the brake microcontroller collapsed due to high fatigue. With the competitions in less than two months, the electronics issues of the car made it unable to complete any official competition lap. I know what it feels like to fail big time, that's innovation!

My roles

Localization and Mapping Lead
- Developed a FastSLAM algorithm to keep track of the cones as landmarks. Additionaly, modified the A-LOAM (Dense SLAM) source code to remove the dependence on GPS tracking as first solution of the FastSLAM algorithm.
LiDAR filtering
- Using a clever combination of grids and filtering techniques, the car was able to see cones at 30 meters distance with a single Velodyne 32C scan.
Self-calibration algorithm
- Using a YOLO network to detect cones with our two cameras, we could estimate the extrinsic parameters of the LiDAR and two cameras via gradient descent.
AI-Enabled backup pipeline
- If the cameras should fail (or a beam of light blinded them), an camera-independent pipeline was developed to identify the color of cones. A 3d-CNN with >90% accuracy for cones in a 12-meter radius.

This was my first time working with SLAM and it inspired me to develop my own SLAM algorithm filling all the blanks the current open-source State-Of-The-Art offered.

Education

Bachelors on Mathematics

Universitat Politècnica de Catalunya (2017 - 2021)

A pure mathematics background gives me the tools I need to confront any type of problem I can encounter. The abstract skills that it has given me has made me highly versatile to learn things really fast.

I know won’t find harder problems to solve than the ones my professors put on our exams, so good news I graduated!

Bachelors on Data Science and Engineering

Universitat Politècnica de Catalunya (2018 - 2022)

A strong abstract background is important but Data Science and Engineering has given me all the practical tools I need to express all my thinkings into feasable and tangible products.

Programming history

I have been programming for more than 10 years, I started creating games with (and for) my friends, I moved to webapps and finally serious projects like an autonomous racing car.

This is me in 2014 in a C++ summer camp in Barcelona, that’s 7 years ago!

This is me and my friends when we made an app to help the university track covid cases (2020).

This is me and my associated in the Math Society Journal after programming a WhatsApp bot capable of connecting +300 math private tutors to more than +700 clients (2021).

This is me and my formula team in Formula Student Germany, which Tesla is a sponsor of! (2021)

Other (unrelated) projects

Due to keeping this resume short and to the point, I won’t comment on other unrelated projects I also did. If anyone is interested, take a look at my LinkedIn!

These range from organizing talks with relevant student alumni, building a network of +300 student private tutors with +700 customers, building from scratch a student-mantained class notes repository and other minor AI-related projects. I love doing things!