Cooperative Target Observation

In recent experiments we examined the effectiveness of various algorithms which direct mobile UAV agents (called observers) to collectively stay within an "observation range" of as many randomly moving targets as possible. Observers and targets live in a sparse continuous 2D or 3D field in MASON. The cooperative target observation environment is shown in Figure 1 (a 3D version is shown in Figure 2). We used this environment to examine "tunably decentralized" cooperative algorithms, where by changing a parameter we could gradually shift the algorithm from one global decision-making procedure to individual per-agent procedures. We examined two such algorithms for controlling the observers based on K-means clustering and hill-climbing.

When launching the applets below, try playing around with the items under the Model tab.

Figure 1. CTO in 2D. Figure 2. CTO in 3D.