Adinfinitium

Adinfinitium Rover Brain v3 / v3.1 (scripted planner)

Status: Final · License: Apache-2.0 · Domain: lunar south pole (NASA LRO LOLA 5 m)

A scripted rover brain with no learned parameters that rolls 36+ candidate plans per rover inside its own world model and follows the best one, plus a gated self-learning layer. This is the rover model Adinfinitium presents. This repo holds the deployed configuration files, not weights.

Model details

Task Rover navigation to ice and return to the lander under power/thermal constraints
Parameters 0 learned (v3: 5 KB config · v3.1: 11 KB config)
Method Planner rolls candidate plans to the end of the episode (waits at the lander, charge or not, three baseline planners, safe harbour); hour-by-hour executor; Dream-RSI layer adapts belief map, belief solar output, energy margin and charge thresholds
v3.1 additions stay-alive route search, 31 new candidate plans, strict-first choice rule, waypoint executor, two new self-learning proposal types (light shifted in time; battery capacity)
Files rover_brain_v3_config.json, rover_brain_v31_config.json
Depends on lunar-hazard-unet for the hazard map

Benchmarks

Four untouched sites (LM2–LM5) × 4 dates, 8,192 runs per method, strict scoring (an empty battery anywhere counts as a lost rover).

Rover brain v3.1 vs baselines

Method Returned Lost
Rover brain v3.1 33.7% 53.9%
Naive planner 17.9% 81.8%
Learned PPO driver (11.4 M, deprecated) 0.0% – (failed)
Learned imitation driver (deprecated) 1.2% – (failed)

v3.1 versus v3 on the same sites: lost −7.1 points [−13.8, −3.5], lower at 4 of 4 sites (v3: 31.0% returned, 61.0% lost). On six fresh sites, v3 under the simulator's lenient rule returned 59.6% with 0.0% lost; under the strict rule 42.0% / 56.7%.

Limitations

  • v3.1 still loses more than half its rovers; of the 583 rovers it saved, 419 only stayed alive on lit ground without completing a trip.
  • The v3 "lost none" result came from a simulator rule that forgives an empty battery within 20 m of the lander.
  • With a wrong belief it loses more rovers than v3 under the lenient rule (late forecast 32.2% vs 19.9%); the four v3.1 test sites are neighbours, so the interval is rough.
  • Not a learned driver: do not describe it as one.

Usage

# project code: https://github.com/oasisk/adinfinitium-handoff
python -m worldmodel.rover3.run eval          # stages: eval, changed, dream, dream-all, dream31, dream31-all

Load rover_brain_v31_config.json into Theta(v31=True); frozen v3 is Theta().

Dream-RSI: gated offline self-improvement

Dream-RSI lets a model keep improving after the main training ends, offline and safely:

  1. Find failures: fly or replay many missions and record where the current model fails.
  2. Dream harder: generate and replay those hard cases (failure-boosted scenarios) and retrain a candidate, warm-started from the current model.
  3. Gate: the candidate is scored on a frozen validation exam (fixed seeds, paired with the current model and the physics backup).
    • Lander: accept only if the candidate loses no mission that the current model or the physics backup lands (zero regressions) AND improves total success.
    • Planner and rover: accept only if the frozen validation metric improves by a margin.
  4. Repeat from whichever model won. Rejected candidates are discarded, and the shipped model never gets worse.

Test maps and the final-exam site are never used in Dream-RSI: training uses training maps, and gating uses a separate validation map.

Measured results (all logged rounds)

Dream-RSI rounds accepted per loop

Loop Rounds accepted Effect on its gate metric
Planner (PlannerSetNet) 1 of 5 Validation Brier 0.01543 → 0.01353 (−12.3%)
Lander pilot (Site11 gate, 1,024 paired missions) 0 of 7 Pilot unchanged; every candidate lost 12–25 missions, so the gate held
Rover brain v3.1 (self-learning on stressed worlds) 9 of 36 Blocked route improved at 4/4 sites, late light 3/4, dusty panels 1/4, early light 0/4
Rover brain v3 9 of 72 Blocked route improved at 6/6 sites, dusty panels 2/6
Rover learned driver v2 5 of 38 Returned rate stayed 0.0%

Across all loops, 18 of 42 first rounds on a new problem were accepted, against 6 of 120 later rounds. What this shows: Dream-RSI reliably repairs specific failures (blocked routes, late light, model faults) without ever shipping a regression. In unchanged conditions it rarely finds a safe gain (0 of 32 rover rounds accepted).

Projections after more Dream-RSI (Projected, not measured)

Median after 20 more generations, from the planner's own Dream-RSI log:

  • Planner Brier: 0.0145 → 0.0094 [0.0043, 0.0145]

Scenario only, for models where Dream-RSI has not been run yet (it borrows the improvement rate seen in the planner and rover loops):

  • Hazard fresh F1: 0.757 → 0.779
  • Light error: 0.450° → 0.410°
  • Terrain error: 2.01 → 1.84 m

The lander and rover v3.1 are projected to stay unchanged, because their gates have not accepted anything in unchanged conditions.

Limitations and next steps

  • The zero-regression gate is strict: it also rejected lander candidates that were slightly better overall (best: 991 vs 990 landings).
  • Training rounds were short (4–5 updates) because of the compute budget.
  • Next: harder self-generated scenarios, longer rounds, and a fresh unseen-map exam after each accepted generation.

Intended use and scope

Research, education and benchmarking of lunar landing-site navigation methods. Not for flight, safety-critical or operational landing decisions. All results are simulation or held-out-lunar-site results, not real-mission performance. Models were frozen (SHA-256) before held-out evaluation; each test site was evaluated once.

About Adinfinitium

Adinfinitium is an open-source effort to make lunar landing and surface navigation easier to study. It began as our NASA Space Apps 2026 project, Lunar Landing Site Navigator, and trains on NASA LRO LOLA 5 m terrain of the lunar south pole. The model chain is: hazard network → light network → terrain model → rover brain → lander pilot. See the whole family at https://huggingface.co/Imok282.

Source code, full records and the hand-off package: https://github.com/oasisk/adinfinitium-handoff (checkpoint SHA-256 values: CHECKPOINTS.sha256).

Citation

@misc{adinfinitium2026lunar,
  title  = {Adinfinitium open models for lunar landing-site navigation},
  author = {Adinfinitium},
  year   = {2026},
  howpublished = {Hugging Face, https://huggingface.co/Imok282},
  note   = {NASA Space Apps 2026, Lunar Landing Site Navigator}
}
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support