Master Thesis Husqvarna Forest Quadruped

MASTER THESIS DESCRIPTION
Husqvarna Forest Quadruped.
The next thing in forestry and landscaping.

About Husqvarna
Husqvarna has a very wide range of professional and non-professional products. Husqvarna is perceived as a leading and innovative brand that excels in performance and maneuverability, enhances productivity and provides high levels of user safety.

Given the megatrend of robotization, we want to explore how autonomous platforms can add value, support or even replace humans in specific tasks, both within forestry and professional landscaping.

Master thesis description
In manual forestry, the user (logger, arborist etc.) has to perform a large number of different work tasks, using a wide range of tools. We are scouting solutions that support the user, perform simpler tasks and add value primarily through working in parallel with the user in a semi- or fully autonomous fashion.
The scope and focus for this thesis work is to train a quadruped (four-legged) robot to navigate through typical forest terrain, avoiding obstacles and finding a route between A and B. The challenge is to determine where it’s possible to walk and where it’s not possible. Light growth, undergrowth or shrubberies can be traversed but solid objects cannot.
The robot in question is planned to be either a Boston Dynamics “Spot” or a Unitree Go2. We assume that the robot is capable to keep balance, traverse uneven terrain etc. on the outset to provide a working platform as a baseline for the thesis work. The general task is to teach the robotic platform to maneuver/navigate in a constantly changing and unpredictable terrain.
This master’s project develops a machine learning model for continuous learning and adaptation in a difficult and changing environment. Since the target environment will be difficult to anticipate, the focus is on adaptability and fast decision making that does not rely on significant training.

We assume that the light conditions is normal daylight, and LIDAR can be used as single or complementary mapping device in combination with a camera.

There are several potential Use Cases:

1) “Follow me” – follow the user through the terrain while carrying some form of equipment
2) “Go fetch” – find the way back from the current position to a fix point, perform some action and either return to the original position, or to a new target position given by the user
3) “Walk from A to B” – the generic use case where the robot needs to map it’s surroundings and determine the best way to reach the target point.
4) Additional Use Cases can be defined in dialogue with Husqvarna but those three are the essence of the scope.
The main task of the thesis work is to propose an algorithm/model/strategy for traversing difficult terrain and in shifting conditions in a forest or park environment.

Thesis work objective
The objective is to develop a proof-of-concept robot application that can be demonstrated in a realistic environment.
The initial questions in the project are
1. How does a robot assess / evaluate its surroundings and determine where it’s possible to walk?
2. What type of information needs to be accessed or created in the target environment
3. Which Use Cases are you addressing?
4. What kind of functional architecture is required to implement the UCs
5. How do we verify the functionality and validate the Use Cases?
6. What system input and output shall be defined?
7. …..?

Skills
The successful candidate should possess or be highly motivated to acquire the following skills:
• Essential
 Experience with deep learning frameworks like PyTorch
 Understanding of SLAM and associated techniques for robotic movement
 Interest in deep learning methods for perception.

• Beneficial Knowledge
 Experience with dynamic simulation of 3D-space
 Familiarity with forestry environment and typical tasks

Location of work
The thesis work is preferably done on Husqvarna's main site in Huskvarna, Sweden.