China
Africa
Explore chapters and articles related to this topic
Embodied AI, or the tale of taming the fungus eater
Published in Arkapravo Bhaumik, From AI to Robotics, 2018
To support the new architectural paradigm of behaviour-based robotics Brooks designed his robotic creatures at MIT using a layering of behaviours, where each behaviour was designed as a finite state model. Locomotion was fundamental to such creatures, with obstacle avoidance the most primitive behaviour, followed by wandering as the second layer, and motion to goal as the third layer. Each of these behavioural layers is a complete loop, from sensor reading to actuation, and mutual interaction between these layers is limited to none. However, a higher layer could support, mitigate, interrupt or override a lower one. Brooks called his artificial creatures mobots (mobile robots) and he and his team designed them on the following principles: An artificial creature has to cope accordingly with changes in its dynamic environment.An artificial creature should be robust enough to endure its environment.An artificial creature should be able to negotiate between multiple goals.An artificial creature should demonstrate purpose in its being, and act to modify its world.
A human behavior model of multi-agent attention based on actor–observer switching for asynchronous motion tasks with limited field of view
Published in Advanced Robotics, 2019
Tingting Zhang-Xu, Kolja Kühnlenz
In behavior-based robot systems, robot various robot behaviors are predefined and an online switching strategy is implemented accounting for situational and task-related parameters. More specifically in the context of vision and motion coordination including attentional processes, known approaches treat top-down or bottom-up attention selection, searching for different target objects in different behaviors, as well as attention enabled or inhibited processes. In this context, [31], a recurrent neural network (RNN) is used to learn the sequence of events encountered during navigation and to make predictions for the future. Attention between object recognition and wall-following task behaviors is switched by the top-down prediction made by the RNN. Another visual attention system, VOCUS, is proposed in [15] for object detection and goal-directed search. This system can detect regions of interest in images in an exploration mode with no specified target and can search for a specific target using top-down information obtained from previous training process as well. Moreover, active gaze control for visual SLAM using features detected by an attention system is applied in [32], which supports the system with tracking, re-detection, and exploration behaviors. In [33], a task-driven object-based visual attention model for robot applications is proposed, which involves five components: pre-attentive object-based segmentation, bottom-up still attention, bottom-up motion attention, top-down object-based biasing, and contour-based object representation. Task-specific moving object detection and still object detection are operated based on this model. A highly competent object recognition system on a mobile robot is proposed in [34], which is capable of locating numerous challenging objects amongst distracters. In previous works, a behavior-based approach for combining bottom-up and top-down attention for robot navigation based on relative entropy is proposed [35]. The known behavior-based approaches are based on a switching mechanism to select between predefined behaviors, which include coordination of vision and motion, and are each fitted to a specific task scenario, in part representing behavior of humans. This paper is embedded in the context of behavior-based robotics, with the difference of investigating the specific human behavior and switching mechanism in case of limited field-of-view and asynchronous motion task constraints as e.g. present in overall formation task settings. The proposed behaviors include ‘actor’ and ‘observer’ behaviors, which have not yet been treated by known approaches.