This paper presents autonomous docking of an inhouse built resident Remotely Operated Vehicle (ROV), called Rover ROV, through acoustic guided techniques. A novel cage-type docking station has been developed. The docking station can be placed on a deep-sea lander, taking the Rover ROV to the seafloor. Instead of using vision-based pose estimation techniques and expensive navigation sensors, the Rover ROV docking adopts an ultra-short baseline (USBL) and low-cost inertial sensors to build an adaptive fault-tolerant integrated navigation system. To solve the problem of sonar-based failure positioning, the measurement residuals are exploited to detect measurement faults. Then, an adaptation scheme for estimating the statistical characteristics of noise in real-time is proposed, which can provide robust and smooth positioning results. It is more suitable for a compact and low-cost deep-sea resident ROV. Field experiments have been conducted successfully in the Qiandao Lake and the South China Sea area with a depth of 3000 m, respectively. The experimental results show that the functionality of autonomous docking has been achieved. Under the guidance of the navigation system, the Rover ROV can autonomously and efficiently return to the docking station within a range of 100 m even when the amounts of outliers exist in the acoustic positioning data. These achievements can be applied to current ROVs by an easy retrofit.