The snake that turns walls into a road.

A narrow shaft usually looks like the end of a route. For EELS, the snake-like robot being developed at NASA’s Jet Propulsion Laboratory, its walls could become the route itself. Rotating screws around the body provide traction; bending joints let the machine change shape. The useful question becomes: where can this body find support?

Think of a climber bracing between opposite sides of a chimney. The gap is useful because both walls are within reach. That analogy explains the ambition more clearly than the word “snake.” EELS borrows the long, flexible body, then adds rotating screws around its surface to turn contact into propulsion. The body is therefore part of the reasoning problem, shaping which choices exist before any route is selected.

Treat contact geometry as part of the route. JPL’s current profile describes counter-rotating screws that supply propulsion and grip across ice and loose material. EELS 1.0 is listed at 100 kilograms and 4.4 metres. The relevant abstraction is a body with distributed contact, rather than a point moving across a map.

There is physical hardware behind that idea. JPL tested prototypes on sand, snow and an indoor ice rink. At Canada’s Athabasca Glacier, the September 2022 work lowered a sensor head into a moulin, a shaft formed in glacier ice. That was a sensing test, not a whole robot independently exploring the depths.

The next year brought a more demanding experiment. EELS 1.5 supported its own weight between ice walls and descended about 1.5 metres. Staff had positioned it using ropes. That distance is modest beside an imagined journey into an alien ocean, but it answers a concrete engineering question: can this mechanism make progress while holding itself in place?

Keep the test configurations separate. EELS 1.0 addressed surface mobility; EELS 1.5 addressed vertical mobility. The September 2023 Athabasca campaign demonstrated weight support and approximately 1.5 metres of descent after assisted placement. This validates a bounded climbing behaviour, not autonomous access from the surface through an entire glacier.

JPL describes an autonomy system that uses lidar and stereo cameras to understand its surroundings and choose routes with risk in mind. That is a more demanding problem than following a painted line: the machine must decide which surfaces can support the body it has, in the shape it can adopt.

The glacier trials also exposed limits. The surface prototype planned a descent into a channel, but its reverse exit required manual assistance. Operators managed its tether. These details make the experiment more informative. “Autonomous” describes particular decisions and behaviours; it does not erase the people, cables or interventions around a prototype. A useful reading of any demonstration asks exactly where the machine’s responsibility began and ended.

Separate decision autonomy from physical independence. In the glacier tests, the tether supplied power and communications, and people managed it. Route planning worked in some cases; reverse travel required intervention. A stopped or assisted run reveals a boundary condition. It should remain in the account alongside successful motion.

Enceladus, Saturn’s icy moon, supplied the original motivation: a robot might descend a vent toward the ocean below. EELS remains listed as in development. JPL’s May 2023 announcement explicitly said it was not part of a NASA mission. Earth tests do not establish readiness for that journey.

The immediate lesson travels further than the destination. Robot design can begin with the awkward space itself. A human-shaped machine makes sense in some settings; here, a body that bends and grips suggests different possibilities. Before asking how smart a robot is, ask what its body allows the software to consider.

Mission status is a separate claim from mobility capability. Enceladus is a motivating concept; the current JPL profile says development. Interpret the glacier evidence as progress toward specific mobility requirements. Generalisation to other environments would require further evidence about materials, geometry, sensing, control and the complete operating system.