Focus Period Lund University 2025

Robot Learning

November 3 – December 5, 2025

Robot learning focuses on developing algorithms and methods that enable robots to learn from interactions with their environment, simulated experiences, as well as humans. In robot learning, traditional machine-learning approaches such as supervised learning and reinforcement learning take on new dimensions, while aspects such as machine learning, data efficiency, reliability, generalization, transfer learning, and safety gain increased importance.  

Robot manipulation, a key domain where robot learning is applied, involves physical interaction with objects and the environment, requiring precise control, decision-making capabilities and robust perception from diverse sensing modalities. Similarly, autonomous navigation requires an understanding of the environment, path planning, and real-time decision-making, often in unpredictable and uncertain settings. Key challenges include interpreting complex, dynamic environments, learning from sparse or unstructured data, and transferring skills to new tasks and environments without extensive retraining. Additionally, transferring skills acquired in simulated environments to real-world applications—known as “Sim-to-Real” transfer—requires meticulous fine-tuning for effective skill adaptation. Safety is paramount as robots learn to navigate and manipulate objects without causing harm to people, the environment, or themselves. Ensuring safety during exploration, when robots try different actions to learn, presents significant risks, especially as mistakes can lead to damaging results. Likewise, safely applying learned policies or actions once deployed requires robust, fault-tolerant design.

Robot learning is driven by the need for robots to function effectively in diverse environments, allowing them to adapt to new tasks and interact with various objects and dynamic agents without continuous reprogramming. The growing interest in robot learning stems from its broad potential. In industry, robot learning can enable flexible automation systems capable of handling variable tasks like assembly, packaging, and inspection. Socially, advances in robot learning could facilitate the deployment of general-purpose robots in human-centered environments, such as homes, healthcare facilities, and educational institutions, while enabling specialized robots, like surgical robots, to operate with even greater precision. However, these advancements bring important considerations. The deployment of robots and autonomous systems that can learn and adapt raises questions around worker displacement, privacy, and ethical machine behavior, which must be part of a larger conversation. As robots become increasingly integrated into our daily lives, understanding and advancing robot learning is not just an academic pursuit but a necessity for a sustainable and beneficial integration. 

Robot learning is an interdisciplinary field, drawing on insights from machine learning, robotics, control theory, and neuroscience to develop adaptive and intelligent systems. Collaboration across these fields is vital for advancing robot learning. As we progress toward a future with more autonomous systems, the ELLIIT focus period initiative seeks to review state-of-the-art techniques, address current and future challenges, and forecast the societal and industrial impacts of robot learning.