Thesis Work Robotics R&D Motion Control

Key Skills

Algorithms, Analytical and Problem solving, AWS, Azure, Large Language Models - LLMs, Machine learning techniques, Natural Language Processing (NLP), Python Programming

Job Description

At ABB, we have the clear goal of driving diversity and inclusion across all dimensions: gender, LGBTQ+, abilities, ethnicity and generations. Together, we are embarking on a journey where each and every one of us, individually and collectively, welcomes and celebrates individual differences.

In ABB Robotics R&D Motion Control we have several Master thesis opportunities. The Motion Control department is responsible for a wide range of areas within the robot controller development spanning from modeling, identification, control design, optimization for path planning and numerous other motion control functionalities.

When applying, please specify which one the two Master thesis you are interested in:

Design and Application of a Machine Learning Framework for Real World Systems in an Automation Environment.

Over recent years, machine learning and in particular deep learning based solutions have established themselves as highly effective for a wide range of tasks. However, to deploy them in an automation environment, they must not only run locally with limited compute but also enable continuous improvements and updates over the product’s life cycle.

In this thesis, we will develop a framework addressing both requirements and then leverage it to design and deploy ML-based solutions for multiple tasks. We will begin by specifying a framework for the efficient development, deployment, and maintenance of ML models in embedded environments. To this end, we will focus on defining a clean interface to the ML-model enabling efficient inference, continuous learning, and updates to the model architecture over time. While establishing the framework described above, we will develop ML-based solutions for a real challenge like collision detection or predictive maintenance, using them first as a case-study to derive requirements and then to demonstrate the effectiveness of the resulting framework.

Details:

• Start: asap for a period of 6 months
• 30 ECTS per student
• Suitable for 1-2 students
• On site ABB Robotics (Finnslätten) in Västerås (exceptions can be made)

 Generative AI for the Analysis of Complex Robotics Systems

Since the publication of ChatGPT, the huge potential of large language models and generative AI in general to efficiently solve a wide range of tasks has become evident. Crucially, these models can be trained and fine-tuned on huge data corpora and then leverage the contained information as context when processing a new input.

While the publicly available tools excel at general purpose tasks such as text generation and analyzing problems or errors in popular programming languages, they perform much worse on more specialized tasks which are barely or not at all represented in their training data. To address this issue, a range of techniques around fine-tuning and in-context learning have been proposed, which will form the core of this project. Depending on the candidate’s interests the thesis will focus on the Automated Analysis of System Diagnostics, the Automated Analysis of Controller Errors, a Copilot for RAPID or other.

Details:

• Start: asap for a period of 6 months
• 30 ECTS per student
• Suitable for 1-2 students
• On site ABB Robotics (Finnslätten) in Västerås (exceptions can be made)

Your responsibilities

• Design and Application of a Machine Learning Framework for Real World Systems in an Automation Environment: Investigate the requirements resulting from the objectives above under deployment on an edge platform. Evaluate both open-source tools and commercial solutions against these requirements in a realistic environment. Based on this evaluation, design a proof of concept. Then, implement it in a real-world control system, realizing one or multiple of the concrete tasks. For any concrete task, a range of suitable ML models should be evaluated, motivated by solutions to similar problems discussed in literature. This part of the project will include an in-depth exploration of both the application domain and available data, including feature engineering and data preprocessing. Finally, Write a high-quality technical report, suitable for publication at an international conference.
• Generative AI for the Analysis of Complex Robotics Systems: Review the available data-sources and curate a dataset suitable for training/fine-tuning and/or testing: Investigate the benefits and drawbacks of a range of state-of-the-art fine-tuning and in-context learning techniques (referred to as training henceforth) for the problem at hand and possible adapt them to the project specific needs. Evaluation multiple open-source and possibly also commercial LLMs on the task at hand using the training techniques reviewed before, and finally implement a prototype demonstrating the potential of the technology to solve the tackled task. Write a high-quality technical report, suitable for publication at an international conference.

Your background

• Motivation to solve real world problems using state-of-the-art methods
• Background in computer science, statistics, engineering, or similar
• Good knowledge of Python
• Excellent problem solving skills
• Experience with code development and tools such as git, conda etc.
• Knowledge of AI & ML, ideally LLMs
• Good interpersonal skills