Mapping Multi-robot 2D mapping: We are investigating the use of mobile robot teams to autonomously explore and map abitrary indoor environments. The overall map is assembled from this data using three key technologies: maximum likelihood estimation, manifold representations and loop closure using mutual observation. Maximum likelihood estimation is used to find the best fit between overlapping range scans; i.e., for each scan, we find the pose estimate that minimizes the discrepancy between it and its neighboring scans. In addition, instead of treating the map as a two-dimensional planar structure, we represent it using a two-dimensional manifold. Loop closure is reduced to the problem of identifying and bringing together those points on the manifold that represent the same point in the world. To recognize such points, we make use of the mutual observations made by the robots. This facilitates a number of interesting capabilities, including lazy loop closure. This video clip illustrates the active loop closure process. This is joint work with Prof. Howard (USC). 3D mapping: Our aim is to generate 3D maps over urban environments that are of the order of one square kilometer in area. We have been working in two mapping approaches: point clouds and planar representation. Point clouds generate dense 3D maps that capture features as small as a few centimeters. Planar representations generate compact and efficient 3D maps for representing building structures. This video clip shows a map of the gerontology building at USC campus generated using a Segway RMP robot and a SICK laser rangefinder. Mapping in dynamic environments: We propose an on-line algorithm for simultaneous localization and mapping of dynamic environments. Our algorithm is capable of differentiating static and dynamic parts of the environment and representing them appropriately on the map. Our approach is based on maintaining two occupancy grids. One grid models the static parts of the environment, and the other models the dynamic parts of the environment. The union of the two provides a complete description of the environment over time. We also maintain a third map containing information about static landmarks detected in the environment. These landmarks provide the robot with localization. Stereo-based and mapping and localization: Stereovison mapping and localization utilizes a combination of feature based methods and dense range data to create high fidelity models of the environment and to localize the robot against the model. We use least-squares statistical Point Estimation techniques to solve the SLAM problem.  This approach encompasses many techniques such as Extended Kalman Filtering and Photogrammetric Bundle Adjustment.  We keep a keen eye on real-time online implementation strategies.  Stereo offers several advantages over classical range finding devices such as 3D information and image intensity information. This information is very useful in the context of data association allowing the robot to disambiguate different locations.

	Localization WiFi-based localization: We are exploring the use of WiFi signal strength information as a cue for localizing mobile robots. Equipped with both the signal strength map and model, robots employ a MCL algorithm to localize themselves. This video clip shows a particle filter converging on the correct robot pose (green dots show the WiFi-based estimate). Empirically, we have found that robots may be localized to within 50cm using only WiFi signal strength and odometry. This is joint work with Prof. Howard (USC). Large-scale 2D localization: Our approach is based on Monte Carlo Localization. The filter is used to localize the robot in the environment. After that, the filter runs the backwards to reconstruct the path of the robot. Experimental tests have been performed on the USC campus. A Segway RMP has been used in the experiments making the task a little bit more difficult due to the constant change in the pitch of the robot. Team localization: We are investigating cooperative methods for relative localization of mobile robot teams; this is, methods whereby every robot in the team can estimate the pose of every other robot, relative to itself. Crucially, our chosen method does not require the use GPS, landmarks, or maps of any kind; instead, robots make direct measurements of the relative pose of nearby robots, and broadcast this information to the team as a whole. Each robot processes this information independently to generate an ego-centric estimate for the pose of other robots.