Research program

The research program is organized around explicit evidence gates. The survey, reference mission, model, and baseline will establish the vocabulary and test conditions together; later papers will depend on what those first results support. Planned papers will appear on arXiv first, with code that runs from a clean checkout. Where the work supports it, a university researcher will review assumptions before the model is fixed and may join as a co-author.

The first result is deliberately narrow: on a planned small-rover reference mission built from published parameters, does a planner using coupled battery and thermal state complete more useful work than a fixed-budget scheduler without violating temperature, state-of-charge, runtime, or uncertainty limits?

Both planners receive the same hardware, initial state, environment, and mission horizon. The report will include survival to the target time, useful activity completed, minimum state of charge, minimum and maximum node temperature, solver runtime, deadline misses, and every constraint violation. Parameter sources and uncertainty ranges are part of the reference mission rather than supporting material left outside it.

Some autonomous spacecraft, servicing robots, and planetary rovers are planned with electrical energy and heat rejection treated outside the task-planning state. The coupling between the task planner and the power and thermal subsystems is then carried through design margin. This paper surveys the literature across spacecraft power systems, thermal control, and autonomous planning. It proposes a taxonomy of how energy and thermal state enter the planning loop, or fail to: as post-hoc checks, as static budgets, as scheduled resources, or as dynamic constraints with physics models in the loop. It identifies open problems at each level, catalogs available open-source tooling, and examines whether the fourth level is practical for small missions. It closes with a reference architecture and a benchmark plan.

Small lunar rovers face a roughly fourteen-day night, and a rover without radioisotope heating must manage its stored energy and thermal state through that interval. This paper couples solar generation, battery state, heater loads, and a lumped thermal network for a representative rover at a mid-latitude site. It will compare a coupled planner with a fixed-budget scheduler across pre-night charging and heater duty cycles, reporting useful work, survival, runtime, uncertainty sensitivity, and constraint violations under the same hardware assumptions.

On-orbit inspection and grapple tasks can be constrained by battery depth of discharge and by the thermal ceiling of actuators and avionics during eclipse. This paper will test which model and planner contracts transfer from the lunar reference mission, then evaluate task ordering and duty cycling against a fixed-schedule baseline. It will not assume that the lunar policy or its validation evidence transfers with them.

Transient thermal simulation becomes too slow for a planning loop as node counts grow. This paper will train physics-informed surrogates on trajectories from an open thermal solver and report accuracy, speed, and constraint-violation rates when the surrogate replaces the solver inside the planner. A surrogate advances only if it meets both the agreement tolerance and the planning deadline.

This paper will evaluate whether the lunar reference mission and later transfer cases can be compared through a common metric set, named baselines, and a published execution environment. It will report agreement and any difference observed when a team outside Mindpool reproduces the benchmark from a clean environment. The benchmark advances only when that independent run is complete and its differences are recorded.

This paper will evaluate the planner on a battery, heater, and radiator breadboard in a thermal-vacuum chamber at a partner laboratory, against the specific claim established by the analytical benchmark. It will report observed battery and thermal behavior, planner actions, the relevant limits, and departures from the modeled case. The hardware result advances only when observed battery and thermal behavior and planner actions meet declared tolerances and limits, with the comparison to the modeled case reported. Hardware work begins only after the analytical result and independent run identify the claim the breadboard is meant to challenge.

An analytical result advances when the model passes its reference checks, the baseline comparison uses the same assumptions, and every violation and deadline miss is reported. It holds when the result is sensitive to an unresolved parameter or cannot be reproduced. It stops when the coupled planner offers no measurable advantage under the declared conditions; that negative result is published with the same artifacts.

Candidate directions include physics-informed surrogates for high-power radiators, shared autonomy for teleoperation under delay, an open benchmark for conjunction screening, and link-budget and safety-envelope tooling for power beaming. A candidate becomes an initiative only when it has a bounded research question, a credible baseline, publishable generic inputs, a reviewer or scenario partner, and a capability not already covered by Initiative 01.

If you lead a laboratory in spacecraft thermal control, space robotics, or machine learning for physics, choose the assumption, baseline, or paper you would like to challenge and write to us with its number. Discuss a paper