TL;DR
This paper introduces Sampling Based Motion Planning (SBMP) techniques, specifically RRT* and SBMPO, as robust solutions for complex spacecraft guidance problems like landing on small celestial bodies and observing their surfaces. These methods offer a more autonomous and less computationally intensive approach compared to traditional optimization techniques, paving the way for future deep-space exploration missions.
What Problem Does This Paper Address?
Future space exploration missions are increasingly demanding greater autonomy for spacecraft. This is driven by communication delays with Earth and the inherent complexity of navigating and operating in the vicinity of small celestial bodies. These environments present unique challenges, including irregular shapes, complex gravitational fields, and stringent collision avoidance requirements. Traditional trajectory optimization methods often struggle with these complexities, requiring precise initial guesses and being prone to converging to suboptimal solutions or failing entirely. This paper addresses the need for more robust and autonomous guidance techniques capable of handling these challenging scenarios.
Methods
The paper proposes two main SBMP algorithms:
- Rapidly Exploring Random Trees (RRT*) for Landing: This algorithm is adapted from robotics for the problem of landing a spacecraft on a small body. RRT builds a tree of possible states and trajectories by randomly sampling the state space. It's designed to efficiently explore complex, high-dimensional spaces and find feasible paths that respect kinodynamic and path constraints. The "star" variant (RRT) ensures that the algorithm converges towards an optimal solution as more samples are added. A polynomial guidance algorithm is used as a "steering method" to generate local optimal trajectories between sampled states.
- Sampling Based Model Predictive Optimization (SBMPO) for Observation: This algorithm is tailored for the "high-level" task of observing the entire surface of a small celestial body. SBMPO builds a 3D map of "observation scores" within the maneuver command space. It iteratively refines this map by sampling maneuvers, propagating trajectories, and evaluating their effectiveness in achieving the observation goal while considering factors like illumination and desired proximity. A weighting factor is used to balance observation score and fuel consumption when selecting the optimal maneuver.
Datasets
Not discussed in the paper. The paper uses simulated mission scenarios:
- Rosetta landing on comet 67P/Churyumov-Gerasimenko: Used to test the RRT* landing algorithm.
- Observation of Didymain in the Didymos binary asteroid system: Used to test the SBMPO observation algorithm.
Key Insights and Ideas
- SBMP for Spacecraft Guidance: The paper demonstrates that SBMP techniques, commonly used in terrestrial robotics, are highly applicable and effective for complex spacecraft guidance problems.
- Handling High-Level Objectives: SBMPO proves capable of translating abstract "high-level" mission goals (like "observe the entire surface") into concrete trajectory plans, a significant advantage over methods requiring explicit waypoints.
- Robustness to Complexity: SBMP algorithms show remarkable robustness in dealing with non-convex constraints and complex dynamics, outperforming traditional optimization methods in terms of reliability and computational time for these specific problems.
- Autonomous Operation: The algorithms are designed to operate with minimal human intervention, a crucial step towards enabling truly autonomous spacecraft missions.
- Efficient Exploration: RRT* and SBMPO efficiently explore the vast state and control spaces, quickly finding feasible solutions even in challenging environments.
Key Results
- Landing Scenario: The RRT algorithm successfully generated feasible descent trajectories for a Rosetta-like landing scenario, demonstrating robustness to the complex shape of comet 67P. While the best solution cost was comparable to direct optimization (2.98 m/s vs. 3.10 m/s), RRT achieved this in significantly less time (seconds vs. minutes) and without requiring an initial guess.
- Observation Scenario: The SBMPO algorithm successfully planned a trajectory to achieve 100% mission completion for observing the surface of Didymain, respecting illumination and collision avoidance constraints. The example showed a spacecraft trajectory that maximized observation time in favorable conditions.
- Computational Efficiency: For the landing problem, RRT* found feasible solutions within 10 seconds and reasonably optimal solutions within 20 seconds on a common PC.
Limitations and Future Work
- On-board Implementation: The authors acknowledge that SBMP algorithms are not yet ready for on-board spacecraft processors due to their computational load. However, their robustness and speed make them promising for future autonomous missions.
- Optimality of Steering Method: While RRT* converges to an optimal solution, the local optimality of the steering method (polynomial guidance) is an approximation.
- Future Work: The paper suggests that SBMP techniques are well-suited for a wide range of future autonomous missions, including assembly of large space structures, inspection of space stations, and capture of debris. Further research into on-board implementation is implied.
Summary updated