Zezhi Liu

In the field of learning from demonstration (LfD), enabling robots to generalize learned manipulation skills to novel scenarios for long-horizon tasks remains challenging. Specifically, it is still difficult for robots to adapt the learned skills to new environments with different task and motion requirements, especially in long-horizon, multistage scenarios with intricate constraints. This article proposes a novel hierarchical framework, called BT-TL-DMPs, that integrates behavior tree (BT), temporal logic (TL), and dynamical movement primitives (DMPs) to address this problem. Within this framework, signal temporal logic (STL) is employed to formally specify complex, long-horizon task requirements and constraints. These STL specifications are systematically transformed to generate reactive and modular BTs for high-level decision-making task structure. An STL-constrained DMP optimization method is proposed to optimize the DMP forcing term, allowing the learned motion primitives to adapt flexibly while satisfying intricate spatiotemporal requirements and, crucially, preserving the essential dynamics learned from demonstrations. The framework is validated through simulations demonstrating generalization capabilities under various STL constraints and real-world experiments on several long-horizon robotic manipulation tasks, where it significantly improves the success rate from 0.25 to 0.85, compared with the baseline. The results demonstrate that the proposed framework effectively bridges the symbolic-motion gap, enabling more reliable and generalizable autonomous manipulation for complex robotic tasks.

World models are a transformative paradigm in embodied AI, enabling agents to learn efficiently and plan by simulating environmental dynamics. As vision-language action and navigation systems grow more sophisticated, integrating world models is crucial for bridging multimodal perception, language understanding, and sequential decision-making. However, existing surveys, which focus on application domains or learning paradigms, have overlooked a critical architectural question: how world models should be structurally integrated into these systems. To address this gap, we introduce an integration-centric taxonomy that classifies research into three fundamental architectural paradigms: (1) Modular Architectures, where world models and policies are distinct modules; (2) Sequential Architectures, implementing hierarchical plan-then-execute workflows; and (3) Unified Architectures, which fuse world prediction and action generation into an end-to-end network. Through this framework, we systematically analyze the inherent trade-offs across these paradigms: modular designs excel in interpretability, sequential approaches enable hierarchical reasoning, and unified models achieve tight prediction-control coordination. Based on this analysis, we outline promising research directions and provide architectural principles for effectively integrating world models, ultimately to advance more efficient, interpretable, and generalizable agents.

The method of dynamic movement primitives (DMPs) is a widely used tool for robots to learn and generalize human skills in different situations. In previous studies, researchers have proposed several methods to generalize the learned human skills when avoiding obstacles. However, the assumption of treating the end-effector of robot as a point is not advisable in actual use. In fact, in all conditions, ignoring the volume of the end-effector can easily lead to unexpected collisions, especially when robots are using volumetric tools or carrying items. Moreover, the orientation of the robot end-effector should be also considered when avoiding obstacles. This article proposes a novel DMPs-based framework to generalize the learned skills with shorter extra distances when avoiding obstacles by adding coupling terms in both the position space and the quaternion space. First, the Minkowski sum of obstacle and end-effector is calculated to represent the collision space, which is then utilized to construct Minkowski sum-based coupling terms in DMPs to avoid collisions. After that, quaternion-based DMPs are used to represent the orientation in Cartesian space, which is used to update the appropriate Minkowski sum at each moment. Furthermore, a quaternion coupling term is proposed to regulate the orientation of the robot end-effector, helping to avoid obstacles in different scenarios. With the introduction of two novel coupling terms in DMPs and quaternion-based DMPs, an advanced skill learning and generalization framework based on DMPs is developed. By integrating both position and orientation, this framework costs only nearly half the additional distance in generalization for obstacle avoidance and demonstrates significantly improved performance across various scenarios. Simulation and experiments have been performed using a KUKA LBR IIWA robot and the results have confirmed the validity of the proposed methods.