Roving the Red Planet

Why the machines exploring Mars have had to learn to decide for themselves

A Every vehicle that has crossed the surface of Mars has done so under a constraint that has nothing to do with engineering and cannot be designed away. A radio instruction from Earth takes between four and twenty-four minutes to arrive, depending on where the two planets stand in their orbits, and the reply takes as long again. Nobody, in other words, has ever driven a rover. The best that a team on Earth can do is decide in the morning where the machine ought to be by nightfall, and then wait through the intervening hours to discover whether it got there.

B The first attempt to work within that limitation was correspondingly cautious. Sojourner, delivered to the surface in 1997 inside a lander cushioned by airbags, was roughly the size of a microwave oven and never ventured more than about a dozen metres from the spacecraft that carried it. Its total mileage over eighty-three working days amounted to approximately one hundred metres. The mission was nonetheless counted an emphatic success, because its purpose was never distance: it was to establish that a wheeled machine could survive the descent, the temperature swings and the abrasive dust at all.

C The rovers that followed in 2004, Spirit and Opportunity, were designed for ninety days apiece and treated as expendable. Spirit lasted six years before becoming embedded in soft soil from which it could not be extracted; Opportunity continued for more than fourteen, covering over forty-five kilometres — a distance no machine had been expected to attempt — before a planet-wide dust storm in 2018 starved its solar panels of light and ended the mission. The pair transformed the geological argument about the planet by identifying minerals that form only in the presence of standing water, but they did so slowly, at a pace dictated by human deliberation.

D Their method was called blind driving. Engineers examined the previous day's images, plotted a route across terrain they could see, and uploaded a sequence of movements to be executed without further judgement. Anything the cameras had not captured — a rock beyond the horizon of the last photograph, a slope steeper than it appeared — was simply unaccounted for. The safeguard was brevity. A drive was kept short enough that the machine would still be standing on ground the operators had inspected when the instructions ran out, which meant that most of a day's daylight was spent stationary while Earth thought.

E The cost of that arrangement was not only measured in metres. A rover halted for most of the Martian afternoon is a rover consuming power and instrument time without generating observations, and every additional sol spent crossing ground already understood is a sol unavailable to the outcrop the geologists actually wanted examined. A software capability existed to do better. Autonomous navigation, in which the rover photographs the ground in front of itself, constructs a three-dimensional model of the immediate terrain, identifies obstacles and chooses its own arc around them, had been available in some form since the beginning. It was seldom used, and the reason was arithmetic rather than caution: the onboard computers were so slow that the vehicle had to halt while it processed each pair of images. Progress became a sequence of pauses, and a machine that stops to think every few metres covers less ground in a day than one that is told exactly where to go.

F Curiosity, which arrived in 2012 with a nuclear power source in place of solar panels, began using the system in earnest, but the change of scale came with Perseverance nine years later. Its navigation software runs on a dedicated processor, which allows the analysis to proceed while the wheels are still turning. Across its first Martian year the rover travelled 17.7 kilometres, and the autonomous system assessed the ground for 88 per cent of it. By comparison, the total distance Opportunity ever covered under its own supervision, across fourteen years, was 2.4 kilometres.

G The records that followed measure the difference precisely. Perseverance has completed a single day's drive of 347.7 metres, and has travelled 699.9 metres without any human reviewing the route in advance — a stretch during which the decisions about where to place the wheels were made entirely on Mars. Its accumulated mileage passed a marathon in under half the time Opportunity required for the same distance. None of this involved a faster vehicle; the top speed of these machines remains slower than a walking pace, and always has been.

H Autonomy is nevertheless not independence, and the distinction matters to the people who operate these vehicles. The rover selects its route; it does not select its destination, and it has no opinion about which rock deserves attention. Scientific judgement remains entirely terrestrial. What autonomy has actually altered is the ratio between travelling and working. Time formerly consumed by conservative, heavily supervised movement is now available for the instruments, and the scientific consequence is subtle but substantial: a rover that can reach a distant outcrop in three days rather than three weeks can afford to be curious about features that would previously have been dismissed as too far away to justify the detour. The argument for sending machines rather than people to Mars has always rested on cost and safety. It increasingly rests on patience as well — on the fact that an instrument which need not be brought home can spend a decade examining a hillside no expedition could ever be funded to visit twice.