A wheeled robot needs a mostly continuous surface under its contact patch — roll it toward a gap wider than its wheel radius, or a loose debris pile steeper than its traction limit, and the wheels spin against nothing (a gap) or against gravel with no grip (a loose pile). Both are binary failures: the rover halts and must reverse and re-approach.
wheel_ok = gap_width < wheel_limit AND slope < traction_limit
leg_ok = ∃ foothold within leg reach, for each of 4 feet independently
A legged robot instead needs only a sparse set of individually reachable footholds — each of its four legs searches its own local reach envelope and commits to whichever point looks stable, so it can straddle a gap that would swallow a wheel, or place feet on the few solid rocks in an otherwise loose pile while the rest of the debris shifts underfoot. Progress becomes a per-leg placement problem instead of an all-or-nothing rolling-contact problem, at the cost of a slower, more deliberate gait.
- Gaps — a break in the rubble; the rover halts if the gap exceeds its wheel radius, the legged robot steps clean over it.
- Steep loose piles — debris steeper than the rover's traction limit spins its wheels in place; the legged robot slows down but keeps placing feet on the pile's stable points.
- Loose rock clusters — both robots slow down here, but the wheeled rover loses far more traction than the legged robot's discrete foot placements.
- After 3 stuck attempts at the same obstacle the rover eventually grinds through at a crawl — real rescue-robot operators do exactly this: reverse, re-approach, and force it, burning time the legged platform never loses.
Real-world relevance: this is the core argument for fielding legged platforms (e.g. ANYmal, Spot) alongside tracked/wheeled UGVs on USAR (urban search-and-rescue) missions — collapsed-structure rubble is exactly the irregular, gap-riddled terrain where continuous rolling contact breaks down but discrete footholds don't.