Robotic Path Planning Method Promises Safer On-Orbit Assembly of Giant Space Telescopes

The dream of building enormous space telescopes in orbit rather than launching them whole has moved a step closer to reality. A research team led by Liu Yinnian at the Shanghai Institute of Technical Physics, Chinese Academy of Sciences, has developed a hierarchical path planning method that allows a robotic manipulator to thread a mirror module through a cluttered, dynamic space environment while still hitting every critical assembly waypoint with millimeter-level precision. The work, published in Space: Science & Technology, addresses one of the most stubborn conflicts in space robotics: the tension between following an exact, pre-planned path and deviating from that path to avoid collisions with unexpected obstacles.

The motivation stems from a fundamental constraint of spaceflight. Launch vehicle fairings impose hard limits on the size of any payload that can be sent to orbit, which means a large-aperture primary mirror simply cannot be launched as a single integrated piece. Robot-assisted on-orbit assembly has therefore become a critical pathway to overcome this bottleneck and enable the large-aperture optical detection systems that increasingly underpin high-resolution remote sensing, the digital economy, and intelligent industries. Yet the assembly environment is unforgiving. A manipulator working around a partially built optical system must simultaneously replicate standardized paths, avoid non-cooperative obstacles that may themselves be moving, and pass precisely through key waypoints where sub-mirror modules must be joined.

Existing approaches have struggled to satisfy all of these demands at once. Global planning methods, which search over the entire configuration space to find a complete path, are limited by the real-time performance bottleneck of high-dimensional search; by the time a solution is found, the situation in orbit may have changed. Local methods, by contrast, react quickly to their immediate surroundings but lack global consistency guarantees, meaning the robot can end up with a locally safe trajectory that globally drifts away from the required assembly path. Dynamic Movement Primitives, or DMPs, offer an attractive middle ground because they encode desired motions as compact parametric trajectories, but standard DMPs struggle to reconcile path shape accuracy with safe obstacle avoidance when multiple objectives compete.

The core insight of the new study is that DMPs contain two distinct families of parameters that can be optimized separately. The internal parameters, which include shape weights, stiffness, and damping, determine the geometric characteristics of the trajectory itself. DMPs model desired motions as spring-damper systems with nonlinear forcing terms, generating arbitrarily complex-shaped trajectories through the weighted superposition of Gaussian basis functions. By adjusting the weight coefficients of different basis functions, the shape of the trajectory can be modulated while preserving its overall structure. The external parameters, such as perturbation terms generated by artificial potential fields, are instead employed to respond to changes in the external environment. This separation allows the framework to decouple the two competing objectives, path passing accuracy and obstacle avoidance safety, into an inner and an outer optimization layer.

At the inner level, the team adopted the Policy Improvement with Path Integrals algorithm, known as PI², to optimize the shape parameter weights of the DMPs. The method works by injecting Gaussian noise into the parameter space, generating multiple candidate trajectories, and then updating the weights through probability-weighted averaging based on a cost function. The cost combines waypoint deviation with trajectory smoothness, and it explicitly incorporates interference constraints between already-assembled and to-be-assembled modules. An exponential enhancement-type penalty term reinforces the approximation to assembly waypoints at specific instants, while a gating function restricts parameter updates to the vicinity of the waypoints. This gating mechanism prevents unnecessary distortion in trajectory segments far from the waypoints, maintaining overall path smoothness while accomplishing the critical task of precise waypoint passing.

The validation results for the inner layer are striking. When the predefined waypoints were deliberately displaced from the reference trajectory by a specified distance in three-dimensional space, the absolute position error of the optimized trajectory at the waypoints was only 0.039 meters, corresponding to a normalized error of 0.048. The convergence behavior of the optimization proved equally encouraging: as the number of PI² iterations increased, the combined cost comprising waypoint deviation and trajectory smoothness continuously decreased and then stabilized, indicating favorable convergence characteristics. In other words, the inner layer reliably drives the DMP trajectory to pass through designated positions without degrading the quality of the path elsewhere.

The outer layer tackles obstacle avoidance through Model Predictive Control. Here the external perturbation term of the DMP is treated as the control input of an MPC scheme. By predicting the future clearance between an ellipsoidal envelope surrounding the end-effector and spheres representing obstacles, the scaling factor of the perturbation term is optimized in a rolling-horizon manner. This means the robot continuously re-plans its avoidance response over a short future window, adjusting to the real-time geometry of the scene. The approach achieves safe avoidance of both static and dynamic obstacles, and because the perturbation acts on top of the internally optimized trajectory, the robot can circumvent obstacles while still maintaining waypoint passing accuracy, with each layer corresponding to a different priority requirement.

Simulation and hardware experiments together demonstrate the effectiveness of the joint internal-external optimization. In simulation, the proposed method achieved minimum obstacle avoidance clearances of 0.078 meters in static obstacle scenarios and 0.018 meters in dynamic obstacle scenarios, with waypoint passing errors as low as 0.006 meters. On the Gazebo physical platform, a UR5 manipulator carrying a hexagonal prism sub-mirror module successfully accomplished both obstacle avoidance and waypoint assembly under both static and dynamic obstacle environments, reaching an absolute position error as low as 0.006 meters. When the obstacle moved at a prescribed speed, the minimum clearance remained positive throughout the motion, confirming that the rolling-horizon optimization keeps the end-effector safe even in changing conditions.

Perhaps the most persuasive evidence comes from the comparison against a reference path without internal and external parameter optimization. In that unoptimized case, the end-effector carrying the sub-mirror collided with the obstacle, with the clearance reaching negative values, and also interfered with already-assembled modules. The contrast validates both the necessity and the robustness of the proposed framework: without the two-layer optimization, a trajectory that looks acceptable on paper can fail catastrophically in a realistic assembly scene, damaging hardware that may be impossible to repair in orbit.

Beyond the specific numbers, the study offers a structured and scalable path planning solution for on-orbit autonomous assembly of large-aperture optical inspection systems. By separating the problem into a shape-optimization layer that guarantees geometric fidelity to the assembly path and a reactive layer that guarantees safety margins around obstacles, the framework provides a clear solution structure for multi-objective coordination that can be extended to other assembly tasks and manipulator platforms. For engineers working to enhance the autonomous operation capability of space robotic systems in complex mission environments, the research represents significant engineering application value, bringing the era of robot-built giant telescopes and large orbital observatories measurably closer.

Subject of Research: Hierarchical path planning based on internal and external parameter optimization of dynamic movement primitives for on-orbit robotic assembly of large-aperture space optical systems

Article Title: Path planning for mirror assembly in optical detection systems: internal and external parameter optimization of dynamic movement primitives

Article References: Path planning for mirror assembly in optical detection systems: internal and external parameter optimization of dynamic movement primitives. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: on-orbit assembly, dynamic movement primitives, path planning, space robotics, model predictive control, PI2 algorithm, obstacle avoidance, large-aperture optics, manipulator, waypoint passing, Gazebo simulation, space telescopes