Hierarchical reinforcement learning (HRL) exhibits remarkable potential in addressing large-scale and long-horizon complex tasks. However, a fundamental challenge, which arises from the inherently entangled nature of hierarchical policies, has not been understood well, consequently compromising training stability exploration efficiency HRL. In this article, we propose novel HRL algorithm, high-level model approximation (HLMA), presenting both theoretical foundations practical...

In this paper, we study Reinforcement Learning from Demonstrations (RLfD) that improves the exploration efficiency of (RL) by providing expert demonstrations. Most existing RLfD methods require demonstrations to be perfect and sufficient, which yet is unrealistic meet in practice. To work on imperfect demonstrations, first define an setting for a formal way, then point out previous suffer two issues terms optimality convergence, respectively. Upon theoretical findings have derived, tackle...

The goal of task transfer in reinforcement learning is migrating the action policy an agent to target from source task. Given their successes on robotic planning, current methods mostly rely two requirements: exactlyrelevant expert demonstrations or explicitly-coded cost function task, both which, however, are inconvenient obtain practice. In this paper, we relax these strong conditions by developing a novel framework where preference applied as guidance. particular, alternate following...

Designing and analyzing model-based RL (MBRL) algorithms with guaranteed monotonic improvement has been challenging, mainly due to the interdependence between policy optimization model learning. Existing discrepancy bounds generally ignore impacts of shifts, their corresponding are prone degrade performance by drastic updating. In this work, we first propose a novel general theoretical scheme for non-decreasing guarantee MBRL. Our follow-up derived reveal relationship shifts improvement....

As an important basis of stable grasping, slip detection plays a critical role on improving the operation level robots. In this paper, novel method that combines unsupervised learning and supervised is proposed. The window matching pursuit used to extract features then SVM applied classify events. Superior other methods, proposed has no restriction grasped object can be easily robot hands. addition, slip-tagging based infrared sensor measures relative distance hand platform consisting...

It has been a challenge to learning skills for an agent from long-horizon unannotated demonstrations. Existing approaches like Hierarchical Imitation Learning(HIL) are prone compounding errors or suboptimal solutions. In this paper, we propose Option-GAIL, novel method learn at long horizon. The key idea of Option-GAIL is modeling the task hierarchy by options and train policy via generative adversarial optimization. particular, Expectation-Maximization(EM)-style algorithm: E-step that...

It is an effective way for the robots to learn operation skills from humans. In this paper, we realize a skill learning system based on teleportation transferring human experience robot. Firstly, robotic teleoperation with wearable device developed by controlling motor speed directly. This greatly reduces time delay comparing that point position. Then, rotation invariant dynamical movement primitive method presented skills. Finally, effectiveness of proposed evaluated experiments Baxter The...

The goal of task transfer in reinforcement learning is migrating the action policy an agent to target from source task. Given their successes on robotic planning, current methods mostly rely two requirements: exactly-relevant expert demonstrations or explicitly-coded cost function task, both which, however, are inconvenient obtain practice. In this paper, we relax these strong conditions by developing a novel framework where preference applied as guidance. particular, alternate following...

In this paper, we study Reinforcement Learning from Demonstrations (RLfD) that improves the exploration efficiency of (RL) by providing expert demonstrations. Most existing RLfD methods require demonstrations to be perfect and sufficient, which yet is unrealistic meet in practice. To work on imperfect demonstrations, first define an setting for a formal way, then point out previous suffer two issues terms optimality convergence, respectively. Upon theoretical findings have derived, tackle...

Learning and inference movement is a very challenging problem due to its high dimensionality dependency varied environments or tasks. In this paper, we propose an effective probabilistic method for learning of basic movements. The motion planning formulated as on directed graphic model deep generative used perform from demonstrations. An important characteristic that it flexibly incorporates the task descriptors context information long-term can be combined with dynamic systems robot...

Model Predictive Control (MPC) has shown the great performance of target optimization and constraint satisfaction. However, heavy computation Optimal Problem (OCP) at each triggering instant brings serious delay from state sampling to control signals, which limits applications MPC in resource-limited robot manipulator systems over complicated tasks. In this paper, we propose a novel robust tube-based smooth-MPC strategy for nonlinear planning with disturbances constraints. Based on piecewise...