The automatic completion of multiple peg-in-hole assembly tasks by robots remains a formidable challenge because the traditional control strategies require complex analysis contact model. In this paper, task is formulated as Markov decision process, and model-driven deep deterministic policy gradient algorithm proposed to accomplish through learned without analyzing states. our algorithm, learning process driven simple force controller. addition, feedback exploration strategy ensure that can...

In this paper, we present an overview of robotic peg-in-hole assembly and analyze two main strategies: contact model-based model-free strategies. More specifically, first introduce the model control approaches, including state recognition compliant steps. Additionally, focus on a comprehensive analysis whole system. Second, without process, decompose learning algorithms into subfields: from demonstrations environments (mainly based reinforcement learning). For each subfield, survey landmark...

Hierarchical reinforcement learning (HRL) can learn the decomposed subpolicies corresponding to local state-space; therefore, it is a promising solution complex robotic assembly control tasks with fewer interactions environments. Most existing HRL algorithms often require on-policy learning, where resampling necessary for every training step. In this article, we propose data-efficient via off-policy three main contributions. First, two augmented MDPs (Markov decision processes) are...

Reinforcement learning (RL) has been increasingly used for single peg-in-hole assembly, where assembly skill is learned through interaction with the environment in a manner similar to skills employed by human beings. However, existing RL algorithms are difficult apply multiple because much more complicated requires sufficient exploration, resulting long training time and less data efficiency. To this end, article focuses on how predict use predicted action control improve efficiency of...

Researchers are currently making progress in unifying the entire gait cycle of powered prostheses by using a human-inspired phase variable, but constructing robust variable to more accurately estimate and desired joint trajectories during varying walking speeds remains an open problem. Firstly, this study proposed piecewise monotonic smooth predict from only measured thigh angle. Compared two most widely used methods, angle integral vs. angular rate angle, method addressed drift problem...

Purpose This paper aims to present an optimization algorithm combined with the impedance control strategy optimize robotic dual peg-in-hole assembly task, and reduce time smooth contact forces during process a small number of experiments. Design/methodology/approach Support vector regression is used predict fitness genes in evolutionary algorithm, which can real-world The parameters are defined as genes, evaluate performance selected parameters. Findings learning-based proposed only...

Peg-in-hole is one of the most frequent mating features between parts, where large friction resistance and poor contact situations result in failure part mating. To address this problem, paper proposes a screw insertion method peg-in-hole assembly for axial reduction. First, effect motion on reduction analyzed with point face contact. Second, clearance-fit discussed; works better than conventional linear jamming prevention; reciprocate strategy also investigated to reduce influence axis...

It remains a formidable challenge for traditional control strategies to perform automatic multiple peg-in-hole assembly tasks due the complicated and dynamic contact states. Inspired by that human could generalize learned skills different well, general learning-based algorithm based on deep deterministic policy gradient (DDPG) is proposed. To make robots learn from experience efficiently stably, learning process driven basic knowledge like PD force strategy. achieve fast in real-world tasks,...

Wire and arc additive manufacturing (WAAM) is a promising method for directly parts with complex shapes. However, the accuracy of existing welding parameter planning methods would dramatically decrease when bead geometry changes dynamically due to long-term dependence, strong coupling, hysteresis properties WAAM process. To this end, nonautoregressive dynamic model proposed predict geometry, an adaptive predictive control (aMPC) plan parameters achieve high accuracy. First, in model,...

Contextual policy search methods have demonstrated the potential to acquire robotic skill generalization on trajectory-shaping-based tasks. However, it is still challenging for contact-rich manipulation tasks because contact force regulation, reference trajectory adaptation, and task must be fulfilled simultaneously. To this end, a hierarchical compliance-based contextual (HC-CPS) approach proposed learn compliant skills force, motion, adaptation. Specifically, parameterized...

The optimal policy of a reinforcement learning problem is often discontinuous and non-smooth. I.e., for two states with similar representations, their policies can be significantly different. In this case, representing the entire function approximator (FA) shared parameters all maybe not desirable, as generalization ability sharing makes discontinuous, non-smooth difficult. A common way to solve problem, known Mixture-of-Experts, represent weighted sum multiple components, where different...

In this paper, micro harmonic vibration was applied to the typical titanium alloys clarify their damping performance at cryogenic temperatures. The effects of modulus different frequencies were elaborately analyzed, and crack propagation mechanism discussed. increase internal dislocations improves eventually leads interface cracking, which is positively correlated with frequency. More importantly, β phase aggregated interface, leading cracks transgranular fractures by stress concentration....

Controlling a biped robot to walk stably is challenging task considering its nonlinearity and hybrid dynamics. Reinforcement learning can address these issues by directly mapping the observed states optimal actions that maximize cumulative reward. However, local minima caused unsuitable rewards overestimation of reward impede maximization To increase reward, this paper designs gait based on walking principles, which compensates for unnatural motions. Besides, an Adversarial Twin Delayed Deep...

Additive manufacturing (AM) is a promising technology. Wire-arc AM (WAAM) has great potential in the metal process, and welding based WAAM its unique advantage. Slicing an essential work AM. Currently, most slicing methods are layer-wise, without consideration of workpiece's geometric topological characteristics, causing problems such as stair-case effect, anisotropic property waste support material, resulting inaccuracy finishing part. This paper proposes spiral method, fully conforming to...