Transformer-Based Warm-Starting for Feasible and Optimal Terminal Approach to Tumbling Objects with Space Manipulators
Abstract
Autonomous capture of a tumbling object with a space manipulator requires trajectory generation under real-time onboard constraints. However, coupled bus-arm dynamics, rotating safety geometry, and visibility requirements make the underlying optimization severely nonconvex, hence expensive to solve and sensitive to initialization. This paper develops a causal-transformer warm-start for the computationally dominant attitude-manipulator stage of a two-stage sequential convex programming (SCP) framework. Across 300 held-out scenarios evaluated on embedded hardware, the learned warm-start reduces the iterations of the second-stage SCP by up to 32% and runtime by 16% while maintaining numerical feasibility and matching the final control cost. Furthermore, linear and flow matching action decoders are compared across action-chunk lengths and training dataset sizes, where the flow matching decoder exhibits stronger training-data efficiency than the linear decoder. When SCP is terminated based on feasibility, the runtime is further reduced while the learned prior suppresses the severe high-cost tail observed with heuristic initialization. These results demonstrate that learned warm-starts can reduce onboard optimization cost while steering SCP toward basins containing feasible, low-cost solutions.
Related
- The starting point of a directed geodesic is special
- Cauchy-Type Convolution Integrals as the Starting Point for a First Encounter with Complex Analysis
- Coupling-convexity of a diffusion with respect to its starting position
- On the feasible set of a membrane shell under confinement. Smooth inactive admissible displacements are dense
- Mixing profile for Glauber dynamics of the discrete Gaussian Free Field starting from super-harmonic functions
BibTeX
@misc{takubo2026transformerbased,
title = {{Transformer-Based Warm-Starting for Feasible and Optimal Terminal Approach to Tumbling Objects with Space Manipulators}},
author = {Yuji Takubo and Maximilian Adang and Mac Schwager and Simone D'Amico},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2606.17317},
eprint = {2606.17317},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2606.17317},
}
RIS
TY - UNPB
TI - Transformer-Based Warm-Starting for Feasible and Optimal Terminal Approach to Tumbling Objects with Space Manipulators
AU - Yuji Takubo
AU - Maximilian Adang
AU - Mac Schwager
AU - Simone D'Amico
PY - 2026
DA - 2026/09/18
PB - arXiv
DO - 10.48550/arxiv.2606.17317
UR - https://arxiv.org/abs/2606.17317
AB - Autonomous capture of a tumbling object with a space manipulator requires trajectory generation under real-time onboard constraints. However, coupled bus-arm dynamics, rotating safety geometry, and visibility requirements make the underlying optimization severely nonconvex, hence expensive to solve and sensitive to initialization. This paper develops a causal-transformer warm-start for the computationally dominant attitude-manipulator stage of a two-stage sequential convex programming (SCP) framework. Across 300 held-out scenarios evaluated on embedded hardware, the learned warm-start reduces the iterations of the second-stage SCP by up to 32% and runtime by 16% while maintaining numerical feasibility and matching the final control cost. Furthermore, linear and flow matching action decoders are compared across action-chunk lengths and training dataset sizes, where the flow matching decoder exhibits stronger training-data efficiency than the linear decoder. When SCP is terminated based on feasibility, the runtime is further reduced while the learned prior suppresses the severe high-cost tail observed with heuristic initialization. These results demonstrate that learned warm-starts can reduce onboard optimization cost while steering SCP toward basins containing feasible, low-cost solutions.
ER -