<p>To address the increasing demands for workspace coverage and generality during on-orbit peg-in-hole assembly, a hexapod robot with autonomous mobility and integrated leg-arm functionality is introduced in this paper as a new paradigm for enhancing space operation capabilities. An assembly posture is adopted in which four legs provide attachment and support while the remaining two legs perform manipulation, enabling six degrees of freedom control of the foot-end effectors through coordinated base pose adjustment. In terms of control strategy, a diffusion-based policy is employed to establish an end-to-end vision-to-action mapping model. Images captured by a foot-mounted camera serve as observational input, from which features are extracted using a conditional variational autoencoder. The diffusion model then generates high-dimensional continuous action sequences to drive the robot in completing peg-in-hole insertion tasks. In both simulation environments and ground experiments, we conducted assembly validation tests on various representative peg-in-hole tasks. The results demonstrate that our method achieves over a 20% improvement in average assembly success rate compared with conventional behavior cloning approaches. These experiments further confirm that the proposed framework can efficiently adapt to a wide range of peg-in-hole configurations, exhibiting superior generalization and scalability relative to traditional methods.</p>