Institute for Dexterous Space Robotics
The Institute for Dexterous Space Robotics is an initiative started by NASA in 2006. This university consortium (University of Maryland, Carnegie-Mellon University, and Stanford University) is performing research to accommodate NASA's current and future needs in the field of space robotics. Specifically, Stanford is working on the problem of docking with a tumbling spacecraft. During the first year, the major goal of the project was to verify Stanford's ability to interface with the hardware testbeds available at the University of Maryland. Many of the demonstrations were planned for SCAMP, a small mobile platform equipped with a camera. This robot is operated in the Neutral Buoyancy Research Facility (NBRF), which is part of the Space Systems Lab (SSL) at UMd. Stanford was able to port existing algorithms written for relative pose estimation and closed-loop control to SCAMP. This initial test demonstrated viable communication between Stanford's in-house code and a foreign hardware platform.
SCAMP with rotating target
Further ongoing research is focused on the problem of sensing the relative motion of a tumbling target, especially when the target is undergoing severe motion (i.e. high angular rates). Students in the ARL are exploring different sensing modalities and the associated challenges.
Range Based Measurement
One possible measurement for relative position is range, e.g. the reading from a laser range finder or LIDAR. Scanning LIDAR exist and have been used to do relative positioning, even for on-orbit flight missions. These algorithms generally rely on predefined reflective targets and a cooperative target, meaning the target is controllable and broadcasting its state information. More advanced algorithms have been developed which can sense relative position by correlating a point cloud of range measurements with a model of the target using the well-established iterative closest point (ICP) algorithm. What happens to these algorithms as the tumbling motion of the target gets faster and faster though? The temporal nature of a scanning sensor results in a problem referred to as "data smearing." This simply means that a point cloud of range measurements is no longer an accurate representation of the target geometry because the target was moving throughout the scanning process. Shown below is a Monte Carlo simulation of the effect target motion has on range measurement error. The satellite model used in the simulation had dimensions based on the Hubble Space Telescope.
Plot of Monte Carlo trials for target angular rate versus average beam offset (from equivalent range to stationary target)
This test demonstrated that, for instance, 40 degrees per second of target motion resulted in an average beam error of up to 0.13 m, which is approximately 10-20 times the documented statistical beam error for commercial laser range finders. To address this challenge, a new estimation scheme has been developed that considers each range measurement individually as opposed to using an aggregate measurement, i.e. a point cloud. This eliminates the data smearing problem because the estimator takes into account the small motion between each successive single beam measurement. The estimator leverages knowledge of the target geometry in a particle filter to achieve this improvement. Simulation results for the estimator correctly determining the target motion are shown below.
Convergence of angular velocity estimate based on simulated sensor measurements
Currently, a hardware platform is under development that will allow the performance of this algorithm to be confirmed using actual sensor data. This hardware platform allows fully programmable motion in the three rotational degrees of freedom while still providing visibility of the target. An image of this demonstrator can be seen below.
Timelapse of tumbling target demonstrator
Last modified Wed, 3 Oct, 2012 at 10:15