Spacecraft Rendezvous Trajectory Generation with Modular Constraints via Diffusion Model Composition

Mariko A. Storey-Matsutani; Richard Linares

cs.ROOct 5, 2026

Spacecraft Rendezvous Trajectory Generation with Modular Constraints via Diffusion Model Composition

Authors: Mariko A. Storey-Matsutani, Richard Linares

Organizations: Graduate Student, Department of Aeronautics and Astronautics, Massachusetts Inst. of Technology, Cambridge MA · Associate Professor, Department of Aeronautics and Astronautics, Massachusetts Inst. of Technology, Cambridge MA

Abstract

Emerging mission classes such as on-orbit servicing, satellite inspection, and active debris removal require trajectory design methods that are adaptable to a variety of mission scenarios. We present a diffusion-based trajectory generation approach for rendezvous and proximity operations (RPO) that enables flexible configuration of mission constraints. First, individual energy-based diffusion models are trained to satisfy distinct constraints such as approach cone and sensor line-of-sight from a set of optimized trajectories. Then, at inference time, the learned energy models can be composed with one another, or with an analytically defined energy field, to enforce specific constraint combinations. We validate this framework with the composition of a learned approach cone model and a learned sensor line-of-sight model, as well as a learned approach cone model and synthetic obstacle avoidance model, both of which yield constraint satisfaction rates that are within 1 percentage point of the single-constraint models or higher. These results indicate that our compositional diffusion framework can provide a modular approach to RPO trajectory design and enable reconfiguration for new constraint combinations without requiring model retraining.

Figures & tables

Parameter	Value
Initial position ranges (m)	([45, 135], [-75, 75], [-75, 75])
Final position (m)	(4.48, 0.05, 0.17)
Time steps per trajectory	50
Time of flight range (s)	[100, 1000]
Approach radius $r_{A}$ (m)	30
Approach cone half-angle $\theta_{A}$	$10^{\circ}$

Table 1: Dataset and Constraint Parameters

Architecture
Parameter	Value
Input shape (time $\times$ features $\times$ channels)	$50\times 19\times 1$
Resolution levels	4
Base channels	32
Channel multipliers	$(1,2,4,8)$
Block structure	$2\times$ ( $3\times 3$ Convolution,

Table 2: U-Net Architecture and Training Hyperparameters

Figure 1: 3D plot of Approach Cone-Only and LOS-Only Trajectory Samples

	Approach Cone Constraint		Sensor LOS Constraint
	Satisfaction (%)	Peak Violation ( ∘ )	Satisfaction (%)	Peak Violation ( ∘ )
Learned Models
Approach Cone	$96.1\pm 9.6$	$0.29\pm 0.76$	$24.2\pm 7.7$	$104.77\pm 14.87$
Sensor LOS	$36.9\pm 27.5$	$20.21\pm 14.66$	$99.6\pm 3.0$	$0.12\pm 0.95$
Training Data
Approach Cone	$97.0\pm 4.5$	$0.32\pm 6.69$	$28.2\pm 8.7$	$104.58\pm 13.89$

Table 3: Constraint Satisfaction and Peak Violation Metrics for the Constraint-Specific Models and Training Data (N=100)

Figure 2: Dynamic Feasibility and Control Effort of Cold-Started SCP, Warm-Started SCP, and Pure Diffusion Samples Across Control Configurations (Generated, Velocity-Based Inverse Dynamics, Position-Based Inverse Dynamics)

Figure 3: Total Wall Time and Constraint Satisfaction of Approach Cone and LOS Models Across Cold-Started SCP, Warm-Started SCP, and Pure Diffusion Samples

Figure 4: 3D plot of Product-Composed Trajectory Samples (N=100)

	Approach Cone Constraint		Sensor LOS Constraint
Model	Satisfaction (%)	Peak Violation ( ∘ )	Satisfaction (%)	Peak Violation ( ∘ )
Product	$99.3\pm 5.1$	$0.03\pm 0.21$	$98.6\pm 7.1$	$0.36\pm 1.76$

Table 4: Constraint Satisfaction and Peak Violation Metrics for the Product Composed Model (N=100)

Figure 5: Comparison of Constraint Satisfaction Along Trajectory

Figure 6: Approach Angle and Sensor Line-of-Sight Angle Over Constraint-Active Trajectory Region

Parameter	Value
Obstacle Center $c$ (m)	(80, 0, 0)
Obstacle Radius $R$ (m)	10
Softening Length $\sigma$ (m)	10
Gain $g$	0.5

Table 5: Obstacle Constraint Parameters

Figure 7: Energy Landscape ( $z=0$ , Z-score normalized for visualization) for the Approach Cone Model, Synthetic Obstacle Model, and Negation Composition

	Approach Cone Constraint		Obstacle Avoidance Constraint
Model	Satisfaction (%)	Peak Violation ( ∘ )	Satisfaction (%)	Min Clearance (m)
Obstacle	$96.5\pm 9.5$	$0.21\pm 0.61$	$100.0\pm 0.0$	$22.86\pm 9.58$

Table 6: Constraint Satisfaction, Peak Violation, and Minimum Clearance Metrics for the Obstacle-Avoidance Composed Model (N=100)

Figure 8: Plane view of Negation-Composed Trajectories and Mean Obstacle Distance Over Trajectory Time

Explore similar work

OrbitTAMP: Grounding Language Models for Task and Motion Planning in Spacecraft Rendezvous
Oct 1, 2026Yuji Takubo, Daniele Gammelli, Marco Pavone +1Large Language Model Planning Spacecraft
Stanford University, Stanford, CA 94305, USA · Department of Aeronautics and Astronautics, Stanford University, Stanford, CA 94305, USA · Italian Institute of Artificial Intelligence (AI4I), Turin, Italy +1
Constrained Diffusion for Data-Scarce Orbital Monte Carlo in Constellation Tasking
Oct 7, 2026Omar Ramadan, Sam Siavoshian, Amir Kashif Saeed +3Monte Carlo
Whiting School of Engineering Johns Hopkins University 3400 N. Charles St. Baltimore, MD 21218, USA
AI-Augmented Model Predictive Control for Safe and Adaptive Rendezvous and Proximity Operations
Jul 18, 2026Luca Sportelli, Tyler Barr, Cagri Kilic +1Model Predictive Control Spacecraft
Department of Aerospace Engineering, Embry–Riddle Aeronautical University, Daytona Beach, FL 32114

Oct 1, 2026, cs.ROJ/K move · Enter open · S save

OrbitTAMP: Grounding Language Models for Task and Motion Planning in Spacecraft Rendezvous

Yuji Takubo, Daniele Gammelli, Marco Pavone, Simone D'Amico

Stanford University, Stanford, CA 94305, USA · Department of Aeronautics and Astronautics, Stanford University, Stanford, CA 94305, USA · Italian Institute of Artificial Intelligence (AI4I), Turin, Italy+1

Large Language Model Planning Spacecraft Motion Planning High-Level Source Code Earth

Open paper AskPDF arXiv

Spacecraft rendezvous and proximity operations (RPO) are currently planned through an expertise-intensive process in which engineers translate high-level operational intent into safe, dynamically feasible trajectories, creating a bottleneck to scalable operations. Large language model (LLM)-based agents could offer an intuitive interface for this process, although their outputs are not inherently grounded in orbital dynamics, operational constraints, or the structure of admissible spacecraft maneuvers. To exploit their semantic reasoning while ensuring the generated plan's physical validity, this paper presents a hierarchical framework for spacecraft task-and-motion planning (TAMP) that grounds LLM reasoning in a graph of reusable behaviors and domain-specific planning modules. Within this framework, a pretrained LLM maps a natural-language command to a partial mission specification. The associated planners then resolve unspecified decisions within the admissible operational space. Finally, trajectory optimization converts the completed mission specification into a dynamically feasible trajectory. Numerical experiments demonstrate that this architecture substantially improves intent recovery over direct LLM generation, achieving 98% exact recovery of partial mission specifications across all evaluated splits when backed by frontier LLMs. Additional test-time-compute experiments show that, for a compact 9B model, verifier-guided revision increases exact recovery from 75% to 88%, while broader behavior-plan search independently improves selection among admissible trajectory realizations. Overall, these results establish a scalable and auditable foundation for language-driven agentic planning of spacecraft RPO.

Abstract

Explore similar work

OrbitTAMP: Grounding Language Models for Task and Motion Planning in Spacecraft Rendezvous

Constrained Diffusion for Data-Scarce Orbital Monte Carlo in Constellation Tasking

AI-Augmented Model Predictive Control for Safe and Adaptive Rendezvous and Proximity Operations