Search

Multi-task robot learning holds significant importance in tackling diverse and complex scenarios. However, current approaches are hindered by performance issues and difficulties in collecting training datasets. In this paper, we propose GeRM (Generalist Robotic Model). We utilize offline reinforcement learning to optimize data utilization strategies to learn from both demonstrations and sub-optimal data, thus surpassing the limitations of human demonstrations. Thereafter, we employ a transformer-based VLA network to process multi-modal inputs and output actions. By introducing the Mixture-of-Experts structure, GeRM allows faster inference speed with higher whole model capacity, and thus resolves the issue of limited RL parameters, enhancing model performance in multi-task learning while controlling computational costs. Through a series of experiments, we demonstrate that GeRM outperforms other methods across all tasks, while also validating its efficiency in both training and inference processes. Additionally, we uncover its potential to acquire emergent skills. Additionally, we contribute the QUARD-Auto dataset, collected automatically to support our training approach and foster advancements in multi-task quadruped robot learning. This work presents a new paradigm for reducing the cost of collecting robot data and driving progress in the multi-task learning community. You can reach our project and video through the link: https://songwxuan.github.io/GeRM/ .

Developing robotic intelligent systems that can adapt quickly to unseen wild situations is one of the critical challenges in pursuing autonomous robotics. Although some impressive progress has been made in walking stability and skill learning in the field of legged robots, their ability to fast adaptation is still inferior to that of animals in nature. Animals are born with massive skills needed to survive, and can quickly acquire new ones, by composing fundamental skills with limited experience. Inspired by this, we propose a novel framework, named Robot Skill Graph (RSG) for organizing massive fundamental skills of robots and dexterously reusing them for fast adaptation. Bearing a structure similar to the Knowledge Graph (KG), RSG is composed of massive dynamic behavioral skills instead of static knowledge in KG and enables discovering implicit relations that exist in be-tween of learning context and acquired skills of robots, serving as a starting point for understanding subtle patterns existing in robots' skill learning. Extensive experimental results demonstrate that RSG can provide rational skill inference upon new tasks and environments and enable quadruped robots to adapt to new scenarios and learn new skills rapidly.

Modern applications of robotics typically involve a robot control system with an inner PI (proportional-integral) or PID (proportional-integral-derivative) control loop and an outer user-specified control loop. The existing outer loop controllers, however, do not take into consideration the dynamic effects of robots and their effectiveness relies on the ad hoc assumption that the inner PI or PID control loop is fast enough, and other torque-based control algorithms cannot be implemented in robotics with closed architecture. This paper investigates the adaptive control of robotic systems with an inner/outer loop structure, taking into full account the effects of the dynamics and the system uncertainties, and both the task-space control and joint-space control are considered. We propose a dynamic modularity approach to resolve this issue, and a class of adaptive outer loop control schemes is proposed and their role is to dynamically generate the joint velocity (or position) command for the low-level joint servoing loop. Without relying on the ad hoc assumption that the joint servoing is fast enough or the modification of the low-level joint controller structure, we rigorously show that the proposed outer loop controllers can ensure the stability and convergence of the closed-loop system. We also propose the outer loop versions of several standard joint-space direct/composite adaptive controllers for rigid or flexible-joint robots, and a promising conclusion may be that most torque-based adaptive controllers for robots can be designed to fit the inner/outer loop structure by using the new definition of the joint velocity (or position) command. Simulation results are provided to show the performance of various adaptive outer loop controllers, using a three-DOF (degree-of-freedom) manipulator, and experiment results using the UR10 robotic system are also presented., Comment: This version is mainly for including the experimental results