This arXiv preprint (not yet peer-reviewed) synthesizes discussions from a Keck Institute for Space Studies workshop attended by roughly two dozen experts from NASA JPL, Caltech, and other institutions. Rather than presenting new empirical data from experiments, it is a conceptual roadmap grounded in expert deliberation, with the inherent limitations that such forward-looking white papers carry: technologies like in-situ reconfigurable hardware remain largely unproven in deep-space radiation and thermal environments, and the verification of systems that can rewrite their own goals poses unsolved safety challenges.

The authors contrast Planetary Exploration 2.0 – NASA’s incremental Mars campaign of 22 progressively capable missions over decades – with the impracticality of repeating that playbook beyond Mars. Cruise times of 10–30 years to ocean worlds or the Oort Cloud preclude iterative learning cycles. Their proposed PE 3.0 envisions one or two “radically adaptive” spacecraft that perform both reconnaissance and follow-up hypothesis-driven science within a single mission by evolving their own capabilities.

At the core are software-defined space systems (SDSS) that treat hardware as reconfigurable through software uploads: FPGAs can be rewired mid-mission for new signal processing, instruments can switch modalities, and modular payloads can be repurposed. Layered on top is embodied AI for autonomous science, navigation, and control, allowing the spacecraft to analyze fresh data, reprioritize targets, and even alter high-level objectives without waiting for round-trip light-time commands from Earth.

The paper’s three mission concepts illustrate the shift: a Neptune/Triton “smart flyby” that uses AI to select high-value observations and perhaps deploy secondary probes; an ocean-world explorer that adapts sampling strategies upon detecting biosignatures or cryovolcanic activity; and an Oort-cloud reconnaissance craft that must operate with near-total independence for decades. These ideas extend far beyond incremental improvements to Perseverance’s AEGIS targeting system or the limited autonomy flown on Cassini and New Horizons.

What the source under-emphasizes is the deeper connection to contemporary AI-agency patterns. Just as LLM-based agents on Earth now decompose tasks, critique their own plans, and pursue open-ended goals with minimal supervision, PE 3.0 spacecraft would embody agency in extreme environments. This mirrors the broader transition from narrow AI tools to autonomous scientific actors – a pattern also visible in terrestrial robotics papers on lifelong learning and self-modifying controllers. The preprint correctly flags verification-and-validation difficulties yet stops short of exploring ethical accountability: if an adaptive spacecraft diverts resources or risks itself based on an onboard inference, where does responsibility lie?

Synthesizing the KISS roadmap with NASA’s 2020 Technology Roadmap for Autonomous Systems (which stresses the necessity of onboard decision-making due to communication delays) and the peer-reviewed 2022 Science Robotics paper on Perseverance’s onboard science autonomy reveals a consistent trajectory. Earlier coverage of outer-solar-system missions has largely missed this convergence: traditional reporting still frames exploration as hardware-plus-preprogrammed-software, overlooking how software-defined reconfigurability plus embodied AI fundamentally changes the mission paradigm from execution to discovery-driven agency.

Genuine analysis shows both transformative potential and practical hurdles. A single PE 3.0 mission could compress decades of incremental science into one 15-year cruise, accelerating answers on icy-moon habitability or primordial Kuiper Belt chemistry. Yet success hinges on radiation-hardened computing that supports real-time machine learning, robust simulation-to-reality transfer for unpredictable environments, and new V&V methods for non-deterministic systems – areas where current technology lags. The workshop’s strength is its interdisciplinary breadth; its limitation is the absence of quantitative trade studies or flight demonstrations to bound the claims.

Ultimately, Planetary Exploration 3.0 is not merely a space-systems paper. It is an early signal that the same drive toward greater AI agency reshaping laboratories and factories is now reaching the edges of the solar system, promising a future in which spacecraft become true scientific partners rather than distant, rigidly scripted emissaries.