Rover Traverse Replanning When New Voids Are Detected
The Problem
A lunar-analog rover traversing a simulated Mare Tranquillitatis pit rim at Mauna Kea was on sol 4 of a planned 6-sol excursion when its deployed acoustic sensors revealed a skylight branch approximately 12 meters to the east of the current path, not visible in any pre-traverse imagery. The pre-uploaded traverse plan had another 200 meters of forward motion committed before the next Earth supervisory window. With no onboard replanner capable of folding the new void into the current plan, the rover would roll past the discovery, burn power, and miss the chance to scout it during this excursion. The supervisor on Earth saw the acoustic update 14 minutes later, but by then the window had passed.
This is the common failure mode of pre-uploaded traverse plans when the subsurface environment reveals new features. Perseverance is now 90% autonomous because Earth cannot waypoint every meter, but the autonomy is mostly navigation-reactive, not science-reactive. ENav evaluates roughly 1,700 candidate paths within a 6-meter horizon for safety and efficiency, and JPL's Self-Driving Perseverance coverage shows ENav looks 15 meters ahead for terrain hazards. What neither addresses is onboard science-driven replanning when new subsurface features emerge from passive data.
Historical context matters here. Curiosity debuted autonomous navigation using hazcams, a major step up from MER-era blind drives, and RIMFAX GPR detected subsurface stratigraphy to 15 meters on Perseverance. Each autonomy layer widens the rover's immediate decision horizon but rarely reaches into science-priority replanning. The DARPA Subterranean Challenge NeBula framework was built precisely for GPS-denied underground replanning and shows what's possible when science-driven triggers are first-class onboard inputs.
Replanning matters most for PSR traverses where margins are tighter due to battery-only windows. A PSR excursion has 60-90 minutes of usable battery time, and a science-driven detour during that window has to fit within the remaining margin or be deferred to a future excursion. EchoQuilt's replanner exposes the battery-margin constraint as a first-class input to the trigger evaluation, so detours that would consume the safety margin are flagged for ground review rather than committed autonomously.
The opportunity cost of missed mid-traverse discoveries is high. Each rover sol on Mars or the Moon costs millions of dollars in mission-cost-amortization terms, and a science feature that the rover rolls past without scouting may not be reachable on a future sol if the traverse plan has committed forward motion. Perseverance's 90% autonomy figure represents a major operational efficiency gain, but the autonomy is tuned for navigation speed and safety, not for science-prioritized detour evaluation. EchoQuilt's replanner addresses the gap by making science discovery a first-class trigger event that can request a deviation from the navigation plan, while still respecting the safety envelope that the navigation autonomy enforces. The result is autonomy that captures opportunistic science without sacrificing the safety guarantees that make the underlying navigation autonomy trustworthy in the first place.
The Solution: Quilt-Triggered Replanning
EchoQuilt's replanner runs onboard alongside the traverse executive and consumes the evolving quilt as it stitches. Every new patch updates a local confidence field. When a patch reveals a feature that crosses a predefined scientific trigger threshold (unexpected void, branching skylight, anomalous thermal gradient signature), the replanner fires and proposes a modification to the committed traverse. The proposal is scored against the original mission objectives, power budget, and communication windows. If the score exceeds a commit threshold, the replanner executes; if not, it logs the proposal for Earth-side review in the next supervisory window.
The trigger-based architecture keeps replanning bounded. A rover does not replan every meter; it replans when the quilt surfaces a feature that matters. Science leads define the trigger set during mission planning, so the replanning policy is human-authored even though execution is autonomous. This matches the NeBula framework design which separates mission-level policy from step-level execution. EchoQuilt's triggers are expressed in a declarative schema, so updating the trigger set between sols does not require a full mission replan.
When a replanning event fires, the rover slots the new detour into the existing plan as a patch rather than a rewrite. If the original plan commits to waypoint W5, and the replanner proposes a 20-meter side excursion to scout a newly detected void, the modified plan becomes "original up through W5, side excursion, return to W5, continue to W6". This preservation-of-structure approach keeps the stitched quilt interpretable after the fact: each patch is tagged with which version of the plan produced it, so Earth-side analysts can reconstruct the decision chain even without live contact.
The patch-preservation approach also handles reversals gracefully. If a replan triggers a detour but the detour patches come back showing the feature is less interesting than expected, the rover can rejoin the original plan without losing committed progress. This failure-recovery behavior matters because science-triggered replanning will sometimes chase false positives, and the architecture has to survive those without cascading errors into subsequent sols.
Quilt-triggered replanning depends on rover platform config because platform choice determines which replanning latencies are available onboard. A wheeled rover has different replanning constraints than an EVA-carried sensor or a tethered descent system, and the trigger thresholds need to match the platform's actual response capability rather than assuming a generic profile.
Advanced Tactics
Budget replanning margin into the base traverse plan, not as a surprise. If a 6-sol excursion expects one or two trigger fires based on historical data from similar sites, the plan should carry roughly 10-15% power and time margin for detours without compromising core objectives. Teams that treat replanning as bonus margin end up having to cancel detours late because the base plan consumed the contingency.
Review trigger-fire logs between sols, not only post-campaign. A trigger that fires frequently on benign features is probably over-tuned and needs threshold adjustment; a trigger that never fires is either well-calibrated or insensitive. EchoQuilt's replanner exports trigger-event logs in a format that fits the MSL supratactical review cadence documented for Curiosity, which means trigger tuning becomes part of the weekly science review rather than a post-mortem exercise.
For analog teams running campaigns in Lanzarote or Hawaii before interplanetary deployment, simulate trigger-fire events from recorded quilts of known caves. Playing historical quilts back into the replanner with variable trigger thresholds tells you which triggers are useful at the target site before you commit the rover. This is the cheapest way to tune a replanning policy for a specific analog target, and the tuning transfers into flight ConOps with minimal adaptation.
A cross-domain analogue from terrestrial rescue extends the dynamic-replanning model. Our route cutoff prevention work covers how terrestrial rescue teams face analogous dynamic-replanning needs when collapse events change route accessibility mid-operation. The rescue community has years of operational experience with re-routing under time pressure, and the heuristics they use for prioritizing route options under uncertainty have informed how EchoQuilt's replanner handles trigger-driven detours under their own time constraints.
Treat the trigger schema itself as versioned, mission-specific configuration rather than universal logic. A trigger set tuned for Marius Hills geometry will perform differently at Mare Tranquillitatis or at a Martian Pavonis-class tube, because the local feature distributions differ. EchoQuilt's replanner accepts trigger sets as named configurations with explicit version metadata, so a sol's science return can be traced back to the exact trigger configuration that was active during execution. This versioning matters when the science team revises trigger thresholds between sols based on what the rover has been finding — without versioning, the audit trail becomes ambiguous about whether a feature was missed because it failed the threshold or because the threshold was changed before the relevant patch was processed.
Coordinate replanning policy with the mission-level science prioritization framework. Triggers that fire on features the science team has already deemed lower-priority will still consume rover time and power, even if the patches they generate are eventually downgraded. EchoQuilt's replanner supports a science-priority weight that scales the commit threshold based on the current campaign objective: when the campaign is focused on skylight characterization, skylight-related triggers get lower commit thresholds and other triggers get higher ones. This dynamic prioritization keeps autonomous replanning aligned with the campaign's evolving scientific focus, rather than treating all trigger types as equally important throughout the mission.
Ready to Run Replanning-Capable Traverses?
JPL mission planners and Artemis architects scoping autonomous tube traverses need a mapping tool where scientific discoveries can modify the plan without an Earth supervisory round trip. EchoQuilt's trigger-driven replanner is built for that mission profile, with human-authored trigger schemas and patch-preserving detour logic. Each pilot ships with a versioned trigger configuration sized to Marius Hills, Mare Tranquillitatis, Pavonis Mons, and Arsia Mons feature distributions, a battery-margin constraint module aligned to PSR excursion windows, an MSL-supratactical trigger-event log format that fits the Curiosity weekly science review cadence, and a science-priority weighting layer that scales the commit threshold based on the current campaign objective. Pilot teams shape the trigger schema versioning conventions and the cross-platform commit-threshold defaults that the 2027 reference release will adopt for Perseverance ENav-compatible autonomy stacks.
Priority goes to JPL Cave Rovers research teams scoping autonomous traversal protocols, Artemis architect working groups scoping crewed-precursor robotic scouting, NIAC PIs targeting Moon Diver Axel-class concepts, Perseverance ENav alumni evaluating science-driven replanning extensions, and ESA PANGAEA campaign coordinators preparing rover-based analog deployments at Lanzarote or Hawaiian sites. Join the Waitlist for Planetary Analog Researchers to run the replanner against your analog traverse plans and help scope the trigger library for lunar and Martian target sites.