K. Madhava Krishna Prem K. Kalra
Robots operating in a real-time environment encounter both stationary and moving objects that need to be negotiated using different schemes in general. Motion planning in a dynamic environment entails tracking moving objects and predicting their future positions. However, this requires as a first step the classification of the objects present in the environment as static or dynamic objects, a step that somehow seems to have been overlooked in the literature dealing with navigation in dynamic environments. Presented here are four schemes for perceiving the presence of dynamic objects in the robot’s neighborhood. The first approach incorporates a network architecture that classifies the robot’s experience of the environment in terms of spatio-temporal sensor patterns as an experience of a static or dynamic object. The second method detects motion by observing changes in the map of the environment it builds and updates. The remaining two approaches use a strategy for representing the objects in the environment through clusters; inspecting the characteristics of the clusters reveals the dynamic objects. These methods are denoted as STA (spatio-temporal approach), MBA (model-based approach), and CBAI and CBAII (cluster-based approach I and II), respectively. The methods have been tested in environments that contain multiple dynamic objects amidst static ones and their efficacy established. A brief comparison of these approaches in terms of criteria critical for real-time collision avoidance has also been presented.