Spacecraft inertial parameters estimation using time series clustering and reinforcement learning

TL;DR

The paper addresses the challenge of inertial parameters that change during spacecraft operations by restricting the problem to a finite set of possible inertia tensors associated with mission status. It proposes a data-driven pipeline that (i) applies a common actuation profile to multiple inertia models, (ii) uses time-series clustering with $Soft-DTW$/$DBA$ metrics on attitude rates to identify the correct inertia identity, and (iii) employs reinforcement learning via PPO to optimize the actuation sequence for maximal discriminability. Key contributions include a robust time-series classifier that pairs with an RL-driven actuator optimizer, and empirical validation in a payload deployer scenario showing resilience to noise and disturbances. The approach offers a practical, in-flight parameter estimation alternative to traditional observers for reconfigurable spacecraft and can be extended to more complex structures or in-orbit servicing tasks.

Abstract

This paper presents a machine learning approach to estimate the inertial parameters of a spacecraft in cases when those change during operations, e.g. multiple deployments of payloads, unfolding of appendages and booms, propellant consumption as well as during in-orbit servicing and active debris removal operations. The machine learning approach uses time series clustering together with an optimised actuation sequence generated by reinforcement learning to facilitate distinguishing among different inertial parameter sets. The performance of the proposed strategy is assessed against the case of a multi-satellite deployment system showing that the algorithm is resilient towards common disturbances in such kinds of operations.

Figures (3)

• Figure 1: RL agent mean reward vs training episode
• Figure 2: Classifier accuracy vs process noise
• Figure 3: Classifier accuracy vs measurement noise