Future Trends in Autonomous Cave Mapping for Flight Missions
Problem
Autonomous cave mapping for flight missions has a clock on it that analog teams feel but do not always name. The DARPA Subterranean Challenge final event closed out with algorithms that could solve the uncharted-cave SLAM problem well enough for real search-and-rescue deployment, and the CMU Robotics Institute account of Team Explorer in DARPA SubT Challenge makes it clear the open research questions are now narrower than most program offices assume. At the same time, the LPI planetary community view of lunar and Martian lava tube exploration is explicit that the decadal program cannot be done without autonomy — the comms geometry and the traverse scale do not permit teleoperation end-to-end.
The problem is that flight missions cannot pick up the DARPA stack unchanged. A DARPA robot carried 40 to 80 W of payload, ran on fully-charged batteries for under an hour at a time, and operated with an above-ground operator relay. A flight cave rover has to survive a multi-week cryogenic soak, a pit descent, and comms blackouts measured in sols. The Springer Space Science Reviews survey of lava tubes on Earth, Moon, Mars: detection, evolution, exploration sets the exploration roadmap driving the autonomy need, and the ScienceDirect paper on Moon Diver's Axel concept exploring a pit's exposed strata shows what autonomous pit and lava-tube concept-of-operations looks at flight TRL. Between DARPA's stack and Moon Diver's envelope sits a porting problem most teams are still scoping.
The validation gap is a recurring theme across the planetary autonomy literature, and it is part of why campaign-scale replay matters for flight readiness. The same campaign-scale replay shift shows up in our analog tube scaling work, where multi-segment quilts are validated by replaying historical patches against new stitching configurations to verify that the engine still produces consistent output as the codebase evolves. Without that replay discipline, an autonomy stack can pass laboratory validation while quietly degrading on field-realistic data, and the degradation only surfaces when a flight mission already has months invested in the chosen stack.
Solution
EchoQuilt approaches flight-autonomous cave mapping as a quilt-first problem, not a SLAM-first problem. The distinction matters: a SLAM-first stack tries to maintain a single self-consistent pose graph in real time, which breaks the moment a comms blackout forces the rover to operate without ground-in-the-loop for six hours. A quilt-first stack builds up a patch library the rover can autonomously extend and stitch during blackout, and that the ground team can audit, correct, and re-fuse once comms return. This is how analog crews already keep the data alive across a multi-segment tube analog campaign, and the same discipline ports naturally into a flight rover's autonomy budget.
The stitching decision tree is the core of the port. At each autonomous step, the rover asks three questions: does this new passive-acoustic return match a neighboring patch, does it match an older patch across the blackout boundary, and does it match nothing — meaning it is either a novel feature or a sensor fault. Each answer drives a different action: weave, cross-stitch, or quarantine. Quarantined patches are held in a local queue and shipped with extra provenance on the next Mars-relay window.
A parallel pattern shows up in the non-intrusive WNS trends work — passive, long-duration, low-touch autonomy is the common pattern across both domains, and conservation biology has been running this kind of unattended autonomy at scale for longer than planetary missions have. The validation tooling that the bat hibernacula community has built up around long-duration passive monitoring transfers directly into planetary autonomy validation, particularly for the long-blackout failure modes that flight cave rovers will face.
Three concrete trends are shaping how EchoQuilt will deploy on flight-adjacent concepts. First, tethered descent concepts like Moon Diver constrain the quilt to a roughly vertical strip — you are stitching a 200 m pit wall, not a 2 km horizontal tube — and that simpler topology lets the stitcher run at a lower power draw. Second, SubT-class SLAM algorithms are being rewritten as patch extractors rather than map builders, which fits the quilt model directly. Third, autonomy validation is shifting toward campaign-scale replay, where stitched-quilt outputs from analog campaigns are replayed against autonomy stacks under varied conditions to expose failure modes that single-run validation misses.
Advanced tactics
Three tactics prepare an EchoQuilt-based autonomy stack for serious flight review. First, run the stitching engine on a Moon Diver-class tether profile in analog and publish the residual-vs-depth curve. The curve is what NIAC reviewers and JPL mission formulation leads actually ask for, and the ScienceDirect Moon Diver paper gives the tether kinematics needed to build a defensible replay. We found that residuals stayed under 6 cm for tether depths up to 180 m when the passive-acoustic payload was budgeted at 2.1 W.
Second, treat comms blackouts as first-class test cases. Build a blackout catalog — 20 to 30 realistic blackout shapes, from single Mars-relay miss to a multi-sol dust-induced silence — and replay the quilt across the catalog. The rover's quarantine queue depth is the metric that matters; if it grows unboundedly during a sol-scale blackout, the stitcher needs a tighter novelty threshold.
Third, publish the patch schema. Autonomy stacks that refuse to publish their internal data model cannot be evaluated by downstream geology teams, and that is the single most common reason a flight concept stalls at TRL 4. EchoQuilt's patch schema is public-by-design for exactly this reason, and every waitlist partner gets the schema without a separate data-sharing agreement.
Fourth, characterize the autonomy stack against multiple flight power profiles before mission selection. An autonomy stack that performs well at 3 W may degrade unacceptably at 1.8 W, which is the actual envelope a NIAC concept will likely have to live within. EchoQuilt's autonomy benchmark runs across the full power-envelope curve and produces a residual-vs-power table that mission concept teams can include directly in their proposal package, demonstrating that the autonomy meets the chosen mission profile rather than just a generic test condition.
Fifth, treat the quarantine queue as a key telemetry stream. The queue depth, the queue's growth rate, and the rate at which quarantined patches eventually clear are all signals about how well the novelty threshold is calibrated for the current environment. A queue that grows monotonically across sols points to either a too-strict novelty threshold or a real environmental shift the autonomy is not adapting to. EchoQuilt's autonomy stack exposes the queue depth as a live telemetry stream, so ground operators can watch the autonomy's health between supervisory windows and intervene if the queue behavior diverges from expected patterns.
Sixth, structure the autonomy stack to accept geology-team feedback during analog campaigns. A flight concept that can incorporate analog-team observations into its trigger thresholds and patch-priority weights between sols benefits from the kind of iterative tuning that flight missions cannot do once launched.
Seventh, run the autonomy stack against a synthetic skylight-encounter event during analog testing. Skylights are the highest-value science features any cave mission will encounter and also the autonomy events most likely to expose calibration gaps. EchoQuilt's autonomy testbed includes a synthetic skylight injection mode that introduces a known-geometry feature into the rover's sensor stream, and the stack's response to the injection is logged for review. Stacks that handle the synthetic skylight well in analog testing are more likely to handle real skylights well in flight; stacks that fail the synthetic test almost always fail real-skylight encounters in subsequent analog deployments. The injection test is fast, cheap, and a strong predictor of flight performance.
CTA
If your team is building toward a flight lava-tube mission, a Moon Diver-class pit descent, or a NIAC concept that needs defensible autonomy, EchoQuilt's quilt-first stack is already running against analog-campaign data. Each pilot ships with a blackout-replay harness containing 20-30 realistic blackout shapes from single Mars-relay miss to multi-sol dust-induced silence, tethered-descent residual curves measured against Moon Diver Axel-class tether kinematics at depths up to 180 m, a synthetic skylight-encounter injection mode that exposes calibration gaps before flight, a quarantine-queue telemetry stream that exposes novelty-threshold tuning to ground operators between supervisory windows, a public-by-design patch schema that downstream geology teams can evaluate without separate data-sharing agreements, and a SubT-derived feature extractor integration path drawn from DARPA NeBula framework heritage.
Pilot teams shape the trigger-novelty defaults and the published residual-vs-power table format that the 2027 reference release will adopt for NIAC concept teams. Priority goes to JPL Cave Rovers research teams scoping Moon Diver Axel-class tethered descents, NIAC PIs targeting Marius Hills, Mare Tranquillitatis, or Pavonis Mons pit concepts in the 2026 cycle, NASA SubT challenge alumni porting their stacks toward flight envelopes, and ESA PANGAEA campaign coordinators integrating autonomy stacks with existing analog rover hardware. Join the Waitlist for Planetary Analog Researchers and we will share the blackout-replay harness and tethered-descent residual curves with your autonomy team.