I am the scientific coordinator of the JCJC Project ANR-24-CE33-0753-01 MeRLin: Multi-limbed Robots empowered by whole-body Loco-manipulation.
Summary
The MeRLin project addresses the limitations of current robotics in performing hazardous and strenuous tasks, which are still predominantly carried out by human workers in industries like construction and heavy manufacturing. Such tasks often expose workers to significant safety and health risks, including long-term musculoskeletal disorders. Motivated by the need to create safer working environments and enhance productivity, the project focuses on developing advanced robotics capable of operating in unstructured and dynamic environments.
The project targets the development of a robot-agnostic framework that enables multi-limbed robots to perform complex loco-manipulation tasks by coordinating their entire bodies. It addresses two primary scientific challenges:
- Whole-body planning: Developing strategies for planning long-term, contact-rich motions that consider the robot’s physical interactions with the environment and manipulated objects.
- Whole-body control: Creating robust, fast-reacting controllers that adapt to changing environments and execute planned tasks effectively.
To achieve these goals, MeRLin proposes a novel integration of model-based optimization techniques (e.g., trajectory optimization and Model Predictive Control) with deep reinforcement learning (DRL):
- Model-based techniques provide guarantees of stability and constraint satisfaction, enabling long-term planning of efficient and feasible robot movements.
- Reinforcement learning ensures robustness and adaptability in dynamic and unpredictable environments, leveraging sensor data like RGB-D cameras and proprioceptive feedback.
A synergy of these methods creates a learning-augmented planning and control framework, which also accelerates learning processes and avoids poor local optima.
MeRLin stands as a timely initiative to meet the growing industry interest in humanoid and multi-limbed robots, as evidenced by significant investments in bipedal robotics from companies like Tesla and Agility Robotics.
Loria scientific team
- Enrico Mingo Hoffman (PI)
- Ioannis Tsikelis (Ph.D. candidate)
Results
- E. M. Hoffman, A. Laurenzi and N. G. Tsagarakis, “The Open Stack of Tasks Library: OpenSoT: A Software Dedicated to Hierarchical Whole-Body Control of Robots Subject to Constraints,” in IEEE Robotics & Automation Magazine, [HAL]
