Hydra: A Networked, Collaborative Simulation and Visualization Tool for Single and Multiple UAV-Assisted Search (Hydra)
Ephrat Bitton and Ken Goldberg
We are developing an algorithmic framework and a networked simulation tool to integrate human decision making with distributed, automated resource control.
Hydra allows searchers operating under different prior distributions to share control over a limited number of camera-mounted UAVs, with the goal of locating a hidden subject as quickly as possible.
Our model is characterized by the following tasks:
- Each agent specifies a rectangular subregion of the search domain to sample for more information, based on his individual prior distribution.
- We then determine which images to take in order to satisfy as many requests as possible without violating resource constraints. We formulate this as an iterative optimization problem, where in each iteration we identify the single best image location and reduce the priority of each overlapping image request by a factor of 1 minus its satisfaction (area of intersection over maximum area) with the selected image. See http://www.ocf.berkeley.edu/~ebitton/mfs/mfsapplet.html for a demonstration.
- Optimal path planning is computed and the UAVs are sent to collect the data.
- We then update each agent's posterior distribution using Bayesian filtering techniques.
We iterate this process until either time expires or the subject is located within a pre-specified margin of error.
Figure 1: Distributed human and automated agents collaborate to assign commands and controls to UAVs
Figure 2: Sketch of Hydra networked interface