News

Speaker: Georgia Chalvatzaki (TU Darmstadt)

Abstract

In this talk, we delve into advanced reinforcement learning (RL) strategies tailored for robotic systems, with a focus on leveraging structure to accelerate learning and ensure safety in robotic manipulation. Our vision is to enable robotic systems to operate safely and reliably in unstructured environments while acquiring a broad set of skills. Traditional RL methods often require extensive data to develop effective control policies. To address this, we explore innovative frameworks designed to enhance learning efficiency, safety, and adaptability. We will cover Hybrid RL methods that incorporate reachability priors for faster task learning, Model Predictive Actor-Critic (MoPAC) approaches that minimize model bias, and techniques for safe exploration to prevent collisions. Additionally, we will discuss Domain Randomization via Entropy Maximization (DORAEMON) for improved sim-to-real transfer, and Deep Diffusion Policy Gradient (DDiffPG) for fostering diverse behavioral learning. Collectively, these methodologies push the boundaries of RL, making it more practical and effective for real-world robotic applications.

Speaker bio

Georgia Chalvatzaki is a Full Professor at TU Darmstadt since April 2023. Before that, she was an Assistant Professor since February 2022, and Independent Research Group Leader from March 2021, after getting the renowned Emmy Noether Programme (ENP) fund of the German Research Foundation (DFG). In her research group, PEARL (previously iROSA), Dr. Chalvatzaki and her team propose new methods at the intersection of machine learning and classical robotics, taking the research for embodied AI robotic assistants one step further. The research in PEARL proposes novel methods for combined planning and learning to enable mobile manipulator robots to solve complex tasks in house-like environments, with the human-in-the-loop of the interaction process.