Hark! Who goes there? Concurrent Association of Communication Channels for Multiple Mobile Robots Plamen Ivanov Dylan A. Shell April 28, 2016 Abstract Robots working in teams can benefit from recruiting the help of nearby robots. But, while robots are typically aware of their neighbors’ relative positions through infor- mation sensed locally (e.g., range and bearing), a robot does not necessarily know the network identifiers (IDs) of its neighbors directly from observation. In this work robots use a simple visual gesture, paired with wireless messages, to rapidly and effectively establish a one-to- one association between the relative positions (local, vi- sual IDs) of neighboring robots and their network ad- dresses (global, wireless IDs). We formalize the channel association problem and ex- plore its structure from an information filter perspec- tive. Under an idealized communication model, we in- vestigate two simple probabilistic algorithms and con- tribute analyses of performance in terms of parameters, such as robot density, communication range, and move- ment speed, Branching Processes are used to predict the macroscopic performance of the algorithms, producing models that characterize the channel association behav- ior, given parameters that describe the multi-robot sys- tem. The approach also allows parameters to be fine- tuned when designing a system so that its performance meets some specified threshold. 1 Introduction Whilst operating as part of a team, robots may recruit the help of those around them. One robot, sensing an- other in a useful place, might send the request: “Will the robot to my right help me move this piano?” But, whereas robots locate others with cameras, laser range finders, or other (spatial) sensors, they use specialized communication devices, like Wi-Fi radios, to send mes- sages to (logical) recipients. Since the same robot will have an identifier in the spatial medium that differs from its identifier in the communicative medium, an associa- tion between the two is needed if the message is to target a specific recipient. How does a robot learn the network addresses of those it wishes to communicate with? A common solution is to compile a set of pairs by hand prior to deployment of the robots. Each pair con- nects a network address with the corresponding visual (a) Visual barcode fiducial markers as described in Howard et al. [2006]. (b) The AprilTag markers as described in Olson [2011]. Figure 1: Examples of visual markers used for identifying robots. identifier and remains static once established. Practi- cal implementations typically use visual marker or fidu- cial systems with high saliency. Influential examples in- clude that of Howard et al. [2006], Olson [2011], Garrido- Jurado et al. [2014], the first two are shown in Fig. 1. In this paper, we tackle the question of relating a local, rel- ative view of a robot to an identifier which can be used to address the robot directly. The multi-robot systems we study are composed of anonymous robots which need not be visually distinguishable from one another. Also, the robots in the system need not share any common spatial reference frame. 1 In the preceding examples, the Wi-Fi and camera are two independent means by which information is ex- changed. When robots can produce some visually iden- tifiable sign, the camera can form part of a visual com- munication channel. The visual fiducials in the photos are excellent for communicating presence of a marker, though a pre-arranged gesture has the benefit of being dynamic. A protocol using just such a visual commu- nication channel is given by Dieudonn´ e et al. [2009]. A second, rather more practical example, is that of Batalin and Sukhatme [2002] who employ a behavior called Dance, which constitutes a gesture to send infor- mation visually. We term the problem of making an association be- tween visual and networked identifiers the communica- tion channel association problem, or association problem for short. This work studies simple probabilistic algo- 1 The systems in Fig. 1 also established a shared spatial repre- sentation and the visual markers were helpful in that regard too— the present work is concerned solely with the association problem. 1