Overview

New space applications include topics like space debris removal, docking, in-orbit servicing, or close proximity operations. Such operations have in common that satellites and probes operate in close vicinity to each other and a high level of automation is required. At the same time there is a high demand for safe maneuvers in order to avoid collisions. Related settings also occur in automated production lines with manipulators and mobile robotic platforms. Depending on the task to be performed, the maneuverability requirements can range from low to high. Ensuring collision-freenes is a challenge especially in close proximity operations as the precise geometry of the involved probes needs to be considered. We focus on trajectory optimization aspects and employ the following techniques:

  • Reinforcement Learning (RL) combining optimal control and waypoint generation in the presence of obstacles
  • a bilevel dynamic programming approach for obstacle avoidance
  • optimal control methods
  • collision detection algorithms

 

Application scenarios:

  • docking maneuvers to a tumbling target for space debris removal
  • close proximity operations for satellite inspection, in-orbit servicing, and satellite protection

 

The methods are implemented and tested on our mobile robots in a lab environment. We use the robot operating system ROS2 and the Unreal Engine for visualization.

Docking Maneuver to a Tumbling Target with Robotic Arm using Reinforcement Learning

The video shows a docking maneuver of a service satellite with a mounted robotic arm to a tumbling target. The trajectory is generated by a reinforcement learning approach with collision avoidance, which takes into account the geometry of the involved bodies. The results have been obtained within the dtec.bw project SeRANIS.

Docking Maneuver to a Tumbling Target with Robotic Arm using Dynamic Programming

The video shows a docking maneuver of a service satellite with a mounted robotic arm to a tumbling target. The trajectory is generated by a bilevel dynamic programming approach with collision avoidance, which takes into account the geometry of the involved bodies. The results have been obtained within the dtec.bw project SeRANIS.

Docking Maneuver to a Tumbling Target

As the number of uncontrollable objects in low earth orbit is rising, the thread of collisions and thus the breakdown of working satellites becomes worth analyzing. Consequently, space debris removal becomes an issue. In this study an optimal docking maneuver of a service satellite to an uncontrolled tumbling target is modeled and solved numerically. After deriving the system dynamics, we introduce boundary conditions to ensure a safe and realizable maneuver and a general Bolza type cost functional to incorporate different optimization goals.

Motion detection for tumbling target

The video shows results for the identification of the motion of a tumbling satellite using a 3d camera and neural networks. A least-squares problem is solved to identify parameters describing the tumbling motion. The results have been obtained within the dtec.bw project SeRANIS.

Object Detection with XTION PRO LIVE and KUKA Youbot

This video shows an KUKA Youbot detecting a red ball with a stereo camera module XTION PRO LIVE.

Youbot High Precision Marking

The video shows an application with the KUKA youBot robot. The task is to mark predefined positions automatically with a high precision. The high accuracy is achieved with an external laser based positioning system.

Collision Avoidance with KUKA youBot

Automatic path planning tasks in robotics can be modeled by suitable shortest path problems. A particular challenge is the computation of optimal paths in the presence of obstacles. For the detection of collisions of geometric bodies we use a linear programming approach, which exploits the Lemma of Gale.

Dynamic Robot Interaction

Automatic path planning tasks in robotics can be modeled by suitable optimal control problems. A particular challenge is the computation of optimal paths of robots that interact dynamically. We use multiple phases to model the maneuver, i.e. an approach phase, an interaction phase, and a return phase. This multiphase optimal control problem is then transformed by standard techniques to a single stage optimal control problem, which can be solved by a direct shooting method.