Fuzzy Logic-Driven Variable Time-Scale Prediction-Based Reinforcement Learning for Robotic Multiple Peg-in-Hole Assembly

A Non-Contact Manipulation for Robotic Applications: A Review on Acoustic Levitation

“…Manipulation in contact is regarded as tasks where explicit or implicit control of interaction forces is required. Tasks such as polishing [7], [8], [9], [10] could be done with implicit control of the forces whilst tasks such as peg-in-hole [11], work piece alignment and articulated motion can be performed without any control of the forces under perfect knowledge. Contact manipulation can be categorized into mainly three types; environment shaping, work piece alignment and articulated motion [12].…”

A Survey of Robot Manipulation in Contact

“…With the term workpiece aligment we mean mainly tasks found in industrial assembly, often variations of the classic peg-in-hole; however, similar tasks are often found inside homes, such as plugging in a socket [40] or assembling furniture [41]. There are almost infinite variations of peg-in-hole, starting with the difference in clearance (industrial assembly often requires very low clearances) and ranging to multi-peg-in-hole where a plug which has two or three pegs that have to slide in simultaneously (for example, [42]). Another flavor is dual-arm peg-in-hole; however, in this case often one arm performs most of the motions, both with humans and robots [43].…”

“…Regardless of the long history, a human can still outperform a robot in peg-in-hole tasks in certain metrics, such as generalization, managing surprising situations and uncertainties and really tight clearances [44]; however, there is work to overcome these, such as meta-reinforcement learning for generalization [45] and clearances smaller than the robot's accuracy (6µM ) [17]. There are also more difficult variants, such as multi-peg-in-hole [42], assembly construction [122], hole-in-peg with threaded parts [123] or pegin-hole combined with articulated motions, which include as folding [75] or snapping [76,77] where either more elaborate motions or force over a certain threshold is required to complete the task. Also, most works in the field assume rigid pieces, but there is also work towards the more challenging field of elastic pieces [73].…”

“…In this section we focus solely on the use of reinforcement learning for in-contact manipulation tasks. Types of tasks that have been addressed using RL include pushing [211], manipulation of articulated objects such as door opening [212], tool use [29], as well as peg-in-hole and similar assembly tasks [174,44,42].…”

“…velocity vs force) has been found to be application dependent [216]. RL can also be combined with other control and decision making approaches, including classical control methods such as impedance control [217] and operational space control [69], planning [122], or fuzzy logic [42].…”