8th Annual Conference on Robot Learning
The Conference on Robot Learning (CoRL) is a scientific conference dedicated to sharing and exploring cutting-edge research and innovation at the intersection of robotics and machine learning. The conference attracts a global audience of researchers, practitioners, and industry experts who are eager to present their findings, discuss challenges, and shape the future of robot learning.
At CoRL 2024, held from November 6 to 9, 2024, in Munich, Germany, Thies Oelerich introduced “Language-guided Manipulator Motion Planning with Bounded Task Space,” co-authored with Christian Hartl-Nesic and Andreas Kugi. Their work advocates for the use of language-based robot control, a method that leverages large language models (LLMs) to interpret environmental contexts for robot manipulator guidance. Despite the versatility of LLMs, safety and performance issues often arise, typically manifesting as jerky robot movements. To address this, the team developed a novel modular framework for zero-shot motion planning in manipulation tasks, which does not rely on motion-planning-specific training. This framework integrates an LLM with a vision model to generate Python code that collaborates with a new path planner. This planner constructs a piecewise linear reference path with safety bounds, ensuring secure navigation. Additionally, an optimization-based planner within the BoundMPC framework executes optimal, safe, and collision-free trajectories along this path. The effectiveness of this innovative approach is demonstrated through various everyday manipulation tasks in both simulated and experimental settings.
