Fig 1.
The contact state in circular shaft-hole assembly.
a) Surface contact (b) Point contact (c) Multiple point contact.
Fig 2.
Contact states in square shaft-hole assembly.
(a) Surface contact (b) Multiple surface contact (c) Line contact (d) Multiple line contact.
Table 1.
DH parameters of BA006N robotic arm.
Fig 3.
Schematic diagram of multi-agent axis assembly.
Table 2.
Pseudo-code of DMDDPG algorithm.
Fig 4.
Assembly task flow.
Fig 5.
Shaft hole assembly virtual space.
Fig 6.
The projection distance between the center points of the shaft hole contact surface.
Fig 7.
The vertical distance between the center points of the shaft hole contact surface.
Fig 8.
Shaft hole insertion.
Fig 9.
Robotic arm axle hole assembly model.
Table 3.
Network training hyperparameters.
Fig 10.
(a) Square shaft-hole assembly; (b) Circular shaft-hole assembly.
Fig 11.
Reward curve of DE-MADDPG algorithm in circular shaft-hole and square shaft-hole assembly.
Fig 12.
Simplified DDPG process.
Fig 13.
Simplified DMDDPG process flowchart.
Fig 14.
Cumulative average total reward value of DMDDPG and DDPG.
Fig 15.
Simplified MDDPG process.
Fig 16.
Cumulative average total reward value of DMDDPG and MDDPG.
Table 4.
Comparison of assembly results of three algorithms.
Fig 17.
Test rig for square shaft-hole assembly.
Table 5.
Force/Torque sensor parameters.
Fig 18.
Force variations under different contact states.
Fig 19.
Average reward curves of experimental and simulated training.
Fig 20.
Assembly experiment object of square shaft hole.
Table 6.
Comparative experiments on square shaft-hole assembly.