Decision Making and Motion Planning
At ETH Zurich's Autonomous Systems Lab I do research
on perception and decision-making and motion planning for a variety of autonomous vehicles,
with a focus on micro aerial vehicles (MAVs)—check out our YouTube channel. I am currently serving as vice-coordinator for the H2020 EU project
centered on ground and aerial robots collaboration for precision agriculture.
My research is currently mainly focused on the
Next Generation Vehicle project,
where I work on decision-making under uncertainty and
integrated perception, planning, and control in the context of autonomous driving
in real-world traffic scenarios.
The project is a collaboration of the
laboratories at the University of Michigan,
Ford Motor Company, and State Farm.
We use a fleet of several autonomous vehicles equipped with INS,
LIDAR and cameras, as shown in the video below.
Visit the project's page to learn more.
During my time at the University of Girona, my work focused on underwater robotics.
Currently I help out at PeRL with the PhD work of Ryan Eustice's students
Jeff's work involves path planning under uncertainty for cooperative localization of AUVs
and Stephen's is focused on planning active SLAM for underwater visual inspection tasks.
My PhD work was centered on the problem of covering a surface of interest with a robot while avoiding
obstacles, particularly in underwater scenarios. My thesis addressed 2D coverage,
uncertainty-driven AUV survey planning, and
autonomous inspection of 3D underwater environments using the
GIRONA 500 AUV, equipped with
optical imaging and sonar sensors.
The automated inspection scheme proposed in the thesis
goes beyond conventional AUV surveys which image the ocean floor from an overhead viewpoint.
In contrast, we propose a 3D coverage path planning algorithm that enables the AUV to
image the target structure from fair view points at different depths.
This enables the AUV sensors to capture the full 3D information of the target structure.
Beyond mere off-line planning, our algorithm adapts the inspection path in real time using sonar range
information, dealing with the uncertainty present both in the low-resolution maps used to plan the path
and in the AUV's localization system.
The video below demonstrates the method at sea in two different inspection tasks using the
GIRONA 500 AUV.
First, GIRONA 500 follows an automatically pre-planned path to inspect a concrete block on a breakwater
structure at 5 m depth. Next, the AUV inspects a natural seamount ranging from 28 m down to 40 m depth.
In both tasks the vehicle acquires bathymetry sonar and stereo camera data used to create 3D models
and photomosaics of the inspected structures.
I participated as core developer (2010) and team leader (2011) with the VICOROB-UdG team in the 2010 and 2011 editions of the Student Autonomous Underwater Challenge - Europe (SAUC-E). Our 2010 team built our torpedo-shaped robot Sparus from the ground up for the competition, and took again the challenge in 2011. We were the champions in 2010 and runner-ups in 2011.
Take a look at our SAUC-E 2010 preparation video diary: