Embodied AI for Decentralized GNC of Collaborative Space Manipulators
At the ELIXIR Lab, we are pioneering next-generation approaches to space robotics, with a focus on Embodied AI (EAI) for decentralized GNC of collaborative space manipulators. Space missions are increasingly long, complex, and communication-delayed, making it impractical for robots to rely on direct human supervision. Robots must instead be capable of autonomous self-modeling, adaptation, and cooperation to operate effectively in orbit and on planetary surfaces.
Our work addresses these challenges by embedding intelligence into the physical embodiment and interactions of robotic systems. Rather than relying on rigid, preprogrammed, or centralized control, we develop decentralized frameworks where each robot can reason about its own capabilities, make decisions locally, and coordinate efficiently with its peers under limited communication bandwidth. This paradigm ensures scalability, resilience, and fault tolerance, all of which are critical for long-duration, high-risk missions in uncertain and dynamic environments.
To achieve this vision, our research is organized around the following themes:
We design distributed frameworks where teams of manipulators jointly perform tasks such as satellite servicing, in-orbit assembly, or planetary construction, without relying on a central controller. Our focus is on enabling robust task allocation, cooperative motion planning, and fault-tolerant control, ensuring that the system continues to function even if communication is delayed or individual robots fail.
Robots in space must adapt to wear, impacts, and partial failures while remaining operational. We develop multimodal self-modeling algorithms that allow robots to learn their own body structure and kinematics using visual, tactile, and proprioceptive data. This embodiment-driven self-awareness enables manipulators to understand their capabilities and limitations, reconfigure control strategies, and recover functionality in uncertain conditions.
Effective collaboration requires a shared understanding of the environment. We are advancing sensing and perception methods such as 3D environment reconstruction, tactile exploration, acoustic feedback, and neural scene representation learning. These approaches allow manipulators to interpret complex environments and exchange compact, meaningful information with teammates through Large Language Models (LLMs), enabling robust teamwork under strict bandwidth and latency constraints. LLMs are also used to develop a Retrieval-Augmented Generation (RAG) paradigm for space manipulators’ collaborative planning.
Through these research directions, the ELIXIR Lab is building the foundation for resilient, intelligent, and autonomous collaboration in space robotics. The innovations developed here will not only transform future space exploration but also impact terrestrial domains such as advanced manufacturing, logistics, and healthcare robotics.
Selected Publications
To come