Power-Starved Mapping: EchoQuilt Basics for Flight Concepts
The Sub-Watt Envelope Cave Cartography Has to Live In
A reviewer closing a MatISSE or NIAC Phase II review on a Marius Hills lander concept tends to open the budget spreadsheet before reading the science section. The surface bus budgets 300 W to 2.5 kW for an interplanetary craft, per NASA's Basics of Space Flight onboard systems chapter, and by the time heaters, avionics, comms, and mobility take their cut, the cartography allocation is rarely above 2 W continuous. NASA's SmallSat SoA power subsystems summary shows the same pattern on the smaller end: tight envelopes mean instrument teams compete for milliwatts.
Spinning LiDAR cannot meet that bar. The published duty-cycle floors for mechanical LiDAR sit around 3-5 W for useful point density. Flash LiDAR trims that number, but not to 300 mW. That leaves planetary cave teams with a pattern familiar to NASA's low-power smart sensor work: cartography has to share a power bus with instruments that cannot tolerate being starved. When the team re-opens a review with a cartography concept stating a 1.8 W peak, the oxygen returns to the room. Concepts that cannot fit under that ceiling do not reach the TRL 5 gate described in NASA's technology readiness levels framework.
The constraint sharpens further on lunar concepts targeting permanently shadowed regions or pit interiors where direct solar charging is unavailable. A Marius Hills concept relying on a small primary battery has perhaps 200 to 400 Wh of total mission energy, and a cartography payload that consumes 5 W means the rest of the mission gets one to two days of operations before the bus dies. JPL's Volatiles Investigating Polar Exploration Rover heritage and the Lunar Volatiles Tracking instrument architectures have already converged on sub-Watt sensing as the only viable approach for power-starved subsurface concepts. EchoQuilt was designed against the same constraint, and its 1.8 W peak draw fits inside the budget that polar and pit concepts have to defend at every Phase A review.
The relationship between power and downlink ceiling matters too, because a payload that survives the bus envelope but generates more data than the bandwidth budgets can return is just as failed as a payload that draws too many watts.
EchoQuilt's Quilt Architecture Inside a 1.8 W Ceiling
EchoQuilt's quilt cartography pipeline sits on a low-power DSP that idles near 20 mW. A single MEMS microphone array at 25 kSPS draws roughly 40-120 mW depending on the number of channels, and an inertial package adds another 15-40 mW. During a stitching burst (patch inference, SLAM update, quilt deltas), the DSP rises to 1.2-1.8 W for durations measured in seconds. Because the active burst is short and triggered by acoustic events rather than a duty-cycled schedule, total energy per sol stays under 5 Wh on a rover-scale platform.
The quilt pattern matters here. Instead of producing a continuous point cloud that has to be compressed and deleted, EchoQuilt stitches small geometric patches at the moment of inference and writes them into a running quilt. Each patch carries an estimated uncertainty envelope; low-confidence patches stay soft and are re-stitched when new evidence arrives, while high-confidence patches lock into the quilt and no longer spend energy on re-inference. This matches the pattern used by neuromorphic vision-in-space work where event-based sensors operate at around 250 uW, a reference class that makes a 1.8 W peak look generous.
A subtle benefit of the patch-locking model is that it bounds the rover's local memory footprint. A LiDAR-first rover has to either keep raw points around for re-registration or rely on aggressive thinning, and both options carry risks at the flight-software level. EchoQuilt locks each high-confidence patch into a fixed-size descriptor (roughly 4 KB including uncertainty metadata) and frees the upstream audio buffer immediately. A typical sol of traverse leaves the rover with a quilt under 8 MB, which fits trivially into flight-grade SRAM and survives a soft reboot without complex recovery logic. JPL flight software heritage favors exactly this kind of bounded-memory design, and it is one of the easier wins to put into a Phase B preliminary design review.
A companion post on flight power fidelity walks through map-accuracy-vs-power curves on analog tubes when the rover is held to flight-class power, which gives reviewers a concrete number to anchor the trade.
A second architectural note matters for ISRU-adjacent missions. EchoQuilt's quilt assembly runs on a deterministic event loop with bounded worst-case execution time, which means the cartography payload can be scheduled alongside other instruments on a flight processor without consuming an unbounded share of compute. JPL flight software heritage prefers payloads that publish their compute budget upfront. EchoQuilt's reference build documents 8 ms peak inference latency on a Cobham GR740 class processor at the 1.8 W sensor draw, which gives flight software architects a number they can plan against rather than discover at integration.
Advanced Tactics for Keeping the Quilt Under Budget
Teams writing a flight concept inside a sub-Watt envelope benefit from three moves. First, separate always-on sensing from event-triggered inference. EchoQuilt's microphone array and IMU stay on continuously because their combined draw is under 150 mW, but patch inference runs only when a novelty threshold fires on the acoustic stream. That keeps average power close to idle even when a traverse produces hours of data.
Second, publish the energy-per-patch curve rather than the average power number. Reviewers who have watched LiDAR concepts miss their energy targets are skeptical of mean-power claims; they want the joule cost of producing one mapped patch at a named uncertainty. An EchoQuilt traverse in the Lofthellir Iceland analog typically produces quilt patches at 0.5-2 J per patch depending on acoustic event density, and that measured number is what belongs in the proposal.
Third, share a TRL roadmap that clears the boundary between TRL 4 (lab breadboard under analog conditions) and TRL 5 (breadboard in relevant environment). Named analog campaigns at Corona, Surtshellir, or Mauna Loa cross that boundary in a single field season when the quilt pipeline is ready beforehand. Teams that try to collapse the two steps into one tend to show up at flight review with unvalidated numbers.
A fourth tactic that compounds the previous three is to instrument the power rail directly during analog runs rather than estimating from a current monitor on the bus. A Hall-effect probe on the sensor head's input feed produces a high-rate trace that captures the inrush at every stitching burst and lets the team fit a power model to actual hardware behavior, and biology-side teams running low-footprint protocols in bat hibernacula have produced the most rigorous milliwatt-class power traces we have seen outside flight programs. Lab-bench estimates routinely under-report inrush by a factor of two because they smooth out events that the flight bus sees as transient draws competing with mobility.
EchoQuilt ships a reference power log format that pairs each quilt patch with the joule cost it actually consumed, and reviewers respond well to a TRL 5 dossier that contains this measured trace alongside the inferred patch geometry. Together the four tactics turn a power-starved cave concept from a credibility challenge into a defensible flight proposal.
CTA
EchoQuilt is building a cohort of flight concept teams who need a cartography payload that survives a sub-Watt bus, a bandwidth-starved telemetry mode, an Electra UHF proximity link, and an analog campaign schedule. If you are a NIAC Phase I or Phase II PI, a JPL instrument lead, or an ESA PANGAEA payload designer preparing a Corona or Surtshellir drop, we want to hand-tune the quilt pipeline to your exact power and data envelope. Each pilot ships with a sub-Watt power profile template tuned for a GR740 flight processor, a measured energy-per-patch curve from analog campaigns at Lofthellir or Mauna Loa, an Electra UHF link-budget analysis sized for pit-bottom geometry, and a Hall-effect probe reference power log format that captures inrush at every stitching burst.
Pilot PIs shape the TRL 5 dossier template that the 2027 reference release will adopt, with priority going to NIAC Phase II concepts targeting Marius Hills, Mare Tranquillitatis, or Pavonis Mons pit descents and to ESA PANGAEA payload designers integrating with JPL Cave Rovers research. Join the Waitlist for Planetary Analog Researchers for early access to the reference hardware and the pipeline source. Priority goes to teams with a TRL 5 target in their next 12 months.