Robot learning is a subfield of robotics and machine learning that takes advantage of a collection of algorithms and methodologies to help a robot learn new skills such as manipulation, locomotion, and classification in either a simulated or real-world environment. At AIRLab, we harness the capabilities of Large Language Models (LLMs) and Vision-Language Models (VLMs) to create end-to-end planning and reasoning robotics frameworks: from generating executable behavior trees for high-level task planning to design multimodal agentic systems capable of long-horizon reasoning and complex planning. Our goal is to bridge the gap between a user-provided language description of the task and a practical representation executable on a real robot. Ultimately, we focus on fully open-source and lightweight models that run entirely on-board with no external APIs or cloud infrastructure required, while embracing zero-shot approaches that require no extra examples or prompt engineering by the user.
Robot Learning
Back to: Autonomous Robotics