Docking of Autonomous Vehicles: A Comparison of Model-Independent Guidance Methods
In this Paper, two generic Line-of-Sight (LOS) sensing-based short-range guidance methodologies are presented for the docking of autonomous vehicles. The first method utilizes a passive LOS sensing scheme to provide vehicle corrective motions, while the latter method utilizes active sensing. The novelty of the proposed guidance methodologies is their applicability to situations that do not allow for direct proximity measurements of the vehicle. The objective of both proposed guidance methods is, thus, to successfully minimize the systematic errors of the vehicle, while allowing it to converge to its desired pose within random noise limits. Both techniques were successfully tested via simulations and are discussed herein in terms of convergence rate and accuracy, in addition to the types of localization problems that each method should be used in.
I understand what Gonzalo is saying, but are there actually any false minima in this situation? I'd like to ask if you've considered a fuzzy control approach, and also where you are in terms of a hardware implementation?
Thank you for all your messages and interests.
Just to answer the questions. We do not encounter the local minima problem, there only exists one solution when the PSD offsets are minimized and that is the desired pose. The gradient method continues to provide motion commands until all these offsets are minimized. SInce as long as there are PSD offsets, the vehicle will try to correct its pose.
We are currently investigating other optimization techniques, fuzzy logic, and neural networks to solve the docking problem. Our only concern with some of the methods is that they can be computationally time consuming.
We are also implementing the two methods on a 3 dof docking problem, using planar LOSs.
Thanks
Goldie
Once you've set it up, a Fuzzy Logic Controller shouldn't be too computationally expensive. You can make it into a big look-up table. FLCs are used in many demanding real-time systems.
Have you had any problems with oscillating around the minimum?
Thank you for the advice, we'll have to look into using fuzzy logic in more detail.
At some poses using the passive method we approach the minimum quickly and then converge at a slower rate. We are addressing this problem by first trying to see if we can optimize the gradient weights, via an adaptive technique based on the detector offsets, so that we can minimize this oscillatory behaviour.
Thanks
Goldie
Very interesting work, good paper. I would like to know if you have tried other optimization techniques for the movement of the vehicle apart from Gradient Descent method?
One of the drawbacks of the gradient descent method is that it converges (sometimes) to a local minimum rather than a global minimum, did you encounter with any of this problem?
Gonzalo.