Search for: All records

This paper proposes a nudged particle filter for estimating the pose of a camera mounted on flying robots collecting a video sequence. The nudged particle filter leverages two image-to-pose and pose-to-image neural networks trained in an auto-encoder fashion with a dataset of pose-labeled images. Given an image, the retrieved camera pose using the image-to-pose network serves as a special particle to nudge the set of particles generated from the particle filter while the pose-to-image network serves to compute the likelihoods of each particle. We demonstrate that such a nudging scheme effectively mitigates low likelihood samplings during the particle propagation step. Ellipsoidal confidence tubes are constructed from the set of particles to provide a computationally efficient bound on localization error. When an ellipsoidal tube self-intersects, the probability volume of the intersection can be significantly shrunken using a novel Dempster–Shafer probability mass assignment algorithm. Starting from the intersection, a loop closure procedure is developed to move backward in time to shrink the volumes of the entire ellipsoidal tube. Experimental results using the Georgia Tech Miniature Autonomous Blimp platform are provided to demonstrate the feasibility and effectiveness of the proposed algorithms in providing localization and pose estimation based on monocular vision. 

Miniature autonomous blimps are autonomous lighter-than-air vehicles that offer a variety of benefits over other existing flight platforms. In particular, blimps offer long flight times, soft envelopes that are resilient to collisions, and friendly human-robot interaction opportunities. As such, these platforms are well suited for indoor applications and human-cluttered environments as catastrophic or life-threatening collisions are far less likely. In this abstract, we detail some of our ongoing efforts to enable autonomous behaviors for lighter-than-air platforms through various sensing, actuation, and swarming efforts. 

Intelligent physical systems as embodied cognitive systems must perform high-level reasoning while concurrently managing an underlying control architecture. The link between cognition and control must manage the problem of converting continuous values from the real world to symbolic representations (and back). To generate effective behaviors, reasoning must include a capacity to replan, acquire and update new information, detect and respond to anomalies, and perform various operations on system goals. But, these processes are not independent and need further exploration. This paper examines an agent’s choices when multiple goal operations co-occur and interact, and it establishes a method of choosing between them. We demonstrate the benefits and discuss the trade offs involved with this and show positive results in a dynamic marine search task. 

In aquaculture farming, escaping fish can lead to large economic losses and major local environmental impacts. As such, the careful inspection of fishnets for breaks or holes presents an important problem. In this paper, we extend upon our previous work in the design of an omnidirectional surface vehicle (OSV) for fishnet inspection by incorporating AI (artificial intelligence) planning methods. For large aquaculture sites, closely inspecting the surface of the net may lead to inefficient performance as holes may occur infrequently. We leverage a hierarchical task network planner to construct plans on when to evaluate a net closely and when to evaluate a net at a distance in order to survey the net with a wider range. Simulation results are provided. 

Due to complex oceanic environments, underwater gliders typically must satisfy a variety of environmental conditions in order to complete high level objectives. Underwater navigation, for example, requires that a glider must periodically surface and re-localize in order to ensure adequate progress is being made. Such conditions may be directly encoded in Hierarchical Task Network (HTN) planners to ensure that glider actions are valid over the execution of a plan. However, HTN planners may not be able to find good solutions when actions have uncertain costs, such as when a glider is disturbed by a flow field. We propose a bounded cost HTN planner that leverages a modified potential search method in order to find good navigation plans that satisfy user-defined constraints. Simulation results are presented to validate the approach.