Advanced industrial robotic assembly requires the process parameters to be tuned to achieve high efficiency: short assembly cycle (AC) time and high first-time throughput (FTT) rate. This task is usually undertaken offline because of the difficulties in real-time modeling and the lack of efficient algorithms. This paper proposes a support vector regression (SVR)-enabled method to optimize the assembly process parameters without interrupting the normal production process. To reduce the risk of obtaining a local minimum, we consider the trade-off between exploration and exploitation and propose an adaptive optimization process to balance the production processes and the optimization outcome. The proposed methods have been verified using a typical peg-in-hole robotic assembly process, and the results are compared with design of experiment (DOE) methods and genetic algorithm (GA) method in terms of efficiency and accuracy. The experimental results show that our methods are able to maintain the high FTT rate when it drops below 99%, shorten the average AC time by 3.4%, and reduce the number of assembly trials to find the optimized process parameters by 99.6%.