The key to a power-efficient increase in rover average speed is faster processing – when driving with stereo vision, visual odometry (VO), and autonomous navigation (Autonav), large fractions of time are spent analyzing imagery for each motion increment. The low duty-cycle of actual driving directly impacts overall power efficiency and speed. For instance, a MER rover running these algorithms is in motion only 10% of the total traverse time, but the power draw of the whole system is almost the same throughout. We will show that with the addition of high-speed low-power co-processing avionics, the duty cycle can be increased to 100%, thereby increasing the average traverse rate by an order of magnitude. This computational improvement will enable driving with machine vision and navigation at traverse speeds and power efficiencies now only possible with MER blind driving.
