Table 2.
Details of the components of the diagram of Fig. 1. For those elements where multiple Research Seeds are given, it is because each addresses a different dimension of the problem; in most cases, many open problems remain for each component
| Component | Definition | Research Seeds |
|---|---|---|
| HOW: | Mechanical alignment | |
| Actuator | Active control of motors (e.g., sequence of motor actions for multi-view stereo) | Moravec (1980) |
| Body | Active control of robot body and body part position and pose (e.g., to move robot to a location more advantageous for current task) | Nilsson (1969) |
| HOW: | Priming | |
| Interpretation | Active adaptation of perceptual interpretation system for current task and physical environment (e.g., tune system to be more receptive to recognition of objects and events relevant to current task) | Williams et al. (1977) (spatial) Tsotsos (1977) (spatiotemporal) |
| Sensing | Active adaptation of sensing system (e.g., to tune sensors to be more sensitive to stimuli relevant to current task) | Bajcsy and Rosenthal (1975) |
| HOW: | Sensor alignment | |
| Optical alignment | Active control of the optical elements of a visual sensor (focal length, gain, shutter speed, white balance, etc.) (e.g., accommodation: increases optical power to maintain a clear image on an object as it draws near) | Tenenbaum (1970) |
| Proprioreceptive alignment | Active control of non-contact, non-visual sensors, such as inertial measurement units (e.g., the choice of path along which the IMU moves to measure linear acceleration and rotational velocity) | Early twentieth century, such as rocket stabilization |
| Exteroceptive alignment | Active control of sensors that measure the interaction with objects and environment such as applied forces/torques, friction, and shape (e.g., the choice of contact pattern over time) | Allen and Bajcsy (1985) |
| WHEN: | Temporal Selection | |
| Instant | Active prediction of when an event is expected (e.g., predicting the object movement in a sequence) | Tsotsos et al. (1979) |
| Extent | Active prediction of how long an event is expected (e.g., predicting the temporal extent of movement in an image sequence) | Tsotsos (1980) |
| WHAT: | Scene selection | |
| Sensory field | Active prediction of where in a scene a stimulus relevant to current task may appear (e.g., selection of the subset of an image where a face outline can be found) | Kelly (1971) |
| Fixation | Active prediction of which portion of a real-world scene to view (e.g., indirect object search, where an easy search for a semantically related object might facilitate search for a target object) | Garvey (1976) (indirect search) Moravec (1980) (interest points) Aloimonos et al. (1988) (ill-posed and nonlinear problems can be well-posed and linear for an active observer) Clark and Ferrier (1988) (saliency-guided head control) Burt (1988) (foveal fixations for tracking) Ye and Tsotsos (1995) (fixation selection for visual search) |
| WHERE: | Viewpoint selection | |
| Agent pose | Active selection of agent pose most appropriate for selecting a viewpoint most useful for current task (e.g., moving an agent to a close enough position for viewing a task-related object or event) | Nilsson (1969) |
| Sensor pose | Active selection of the pose of a sensor most appropriate for the current task (includes convergent binocular camera systems) (e.g., pointing a camera at a target in with the best viewing angle for its recognition) | Brown (1990) (general gaze control) Coombs and Brown (1990) (binocular vergence; use of nonvisual cues in stabilizing gaze) Wilkes and Tsotsos (1992) (viewpoint behaviors for recognition) |