Platforms

From bridges to factories, to stop things failing and collapsing routine inspection is needed. However, when the target is halfway up an industrial chimney or inside a boiler the task becomes dangerous, time consuming and expensive. Because of this a large part of our work in enabling autonomous MAVs, has been with the focus of inspecting difficult to access areas. One example of this is in the EuRoC project.

While industrial inspection is useful, there is only so much that can be done as a passive observer. To take the next step MAV systems must make contact with and modify their environment. To do this the MAV must be fitted with actuators. To ensure the system remains stable and can apply a required force in the presence of disturbances we must also look to over-actuated systems that can decouple their position and orientation. This was the focus of the recently finished Aeroworks project.

Current approaches to farming must always consider the time required for a person to perform a task. This means that while assessing the health of plants on an individual basis or only spraying for weeds where they are growing would produce better crops, it is too labor intensive. This results in a farmer performing blanket operations such as harvesting an entire field even though some areas are not fully grown. Robotic systems will allow the control of a farm at the plant level improving yield and quality while reducing the use of herbicides and pesticides. These aims are explored in the European project Flourish.

Aerial systems have the ability to quickly survey  an area that would be difficult or dangerous to cover on foot. This makes them invaluable in searching for individuals outdoors. However, due to the limitations of wireless signals, searching underground or in dangerous / collapsed buildings is not feasible using manual pilots. We aim to develop approaches that will enable MAV systems to autonomously explore such areas to allow an assessment of if a location is safe or to locate missing individuals.

Every year the ASL brings together a group of 8 to 12 talented Bachelor students who work together to create a new type of robotic system. During my work I was fortunate enough to help supervise the Voliro project, which had the aim of creating an Omni-directional aerial vehicle that could still acheive an efficient hover configuration. The project was successful and from it we have created a new area of research in our group looking into overactuated systems and their advantages.

The European Robotics Challenge is an FP7 project that aims to get research out of the lab and into the hands of European industry. To this end it has paired academic groups with industrial partners and provided them with assistance and funding to develop a robotic solution to an identified industrial need. 

The challenge consists of three seperate domains, robotic arms, mobile manipulators and micro aerial vehicles (MAVs). ETH are the hosts of this third domain.

As hosts we have provided the challenges with all the basic libraries needed to get an autonomous MAV up and flying (VI state estimation, trajectory generation, a MPC, dense mapping, etc), set tasks where they must demonstrate their ability to create a robust autonomous system and provided assistance where possible.

The Mohamed Bin Zayer International Robotics Challenge 2017. "The competition is motivated by the technological challenges facing the next generation of robotics which is poised to have a trans-formative impact in a variety of new applications and markets". 
The challenge was centered around three tasks that had to be completed autonomously. 1) landing an MAV on a moving platform, 2) using a UGV to turn a value, 3) pick up and delivery of items using an MAV. A grand challenge combined all of the above tasks at once. ASL in collaboration with RSL managed to claim 2nd place in the grand challenge.

A competition where a ground and flying vehicle have to collaborate to explore an area in a simulated disaster situation. Our aim of near complete autonomy caused us no end of issues at the challenge, but proved a good test for our newest platform (Jay) and our new mapping library (Voxblox)

This competition presented the challenging task of autonomously identifying and flying through a series of hoops in an arena. The hoop specs required us to integrate a new smaller platform hawk, and get it up and running with our flight stack

Image Undistort

What started as a way to quickly visualize the undistortion results from Kalibr has grown into a fully featured undistortion / rectification / stereo vision library. Its aim is to fill a few niches that stereo_image_proc doesn't quite cover.

Voxblox

A voxel hashing truncated and euclidean signed distance field (TSDF/ESDF) library. Unlike almost all other TSDF libraries it can run CPU only while still maintaining real-time performance. It also explicitly maps all free-space and scales well to larger voxels to allow its use in robotic exploration, mapping and planning applications. 

State Monitor

A dead simple tool for quickly visualizing and monitoring the output of many of ASL's state estimation tools. It has been designed to be simple to use and easily extendable to other estimators.

Github Links

For a full list of open source code I have contributed to see my github page: 
https://github.com/ZacharyTaylor

Also see the excellent work done by the rotary wing group and ASL as a whole at: https://github.com/ethz-asl

Misc Matlab Functions

A collection of small matlab scripts, mainly used in processing and projecting 3D point clouds. Written during my PhD