The synthetic forager (SF) project aims to understand the basis of foraging behavior by constructing an autonomous agent that can act in the real world. To achieve this objective we need to construct a perceptual, cognition, behavior system for a synthetic forager. We are currently developing and testing a biologically constrained neuronal model based on the neurophysiological and ethological investigation of foraging and optimal decision making in rodents. Supported by  ICT–217148

Partners:

• SPECS, Universitat Pompeu Fabra
• Tel Aviv University
• IDIBAPS Consorci Institut d’Investigacions Biomèdiques August Pi i Sunyer
• Universiteit van Amsterdam
• Universität Osnabrück
• Guger Technologies OEG
• Robosoft

Visit the Synthetic Forager project’s Youtube video channel for lectures, experiment videos, and a documentary!

Methods

The cognitive Distributed Adaptive Control architecture (DAC [1]) is the framework in which we integrate all building blocks, see Figure 1. DAC is a robot based neuronal model of perception, cognition and behavior that is a standard in the domains of new artificial intelligence and behavior based robotics [2].

Figure 1: Schema of the DAC architecture. The reactive layer endows a behaving system with a prewired repertoire of reflexes (association of unconditioned stimuli with their responses US and UR). The adaptive layer provides the mechanisms for the adaptive classification of sensory events (conditioned stimulus, CS) and the reshaping of responses (conditioned responses, CR) is a model of classic conditioning. The contextual layer describes goal-oriented learning as observed in operant conditioning.

Figure 2. One example of the application of DAC to robot random foraging. In these tasks the robot is freely exploring an environment that contains obstacles and targets.

Perceptual learning system: Sensory processing is a key component of successful foraging. An Hippocampal based navigation system is being developed for this purpose.

Figure 3: The hipocampal based navigation system is based on learning the so called place cells, neurons that fire in a certain place of the environment and that are learned from different sensory modalities.

Rule learning and planning system: In the higher contextual level of DAC, a more complex rule learning system and goal driven planning drives the system. The assessment of value to internal representations and possible actions is a critical step for an optimal foraging agent. Rules and more complex reward sequences are managed and stored in short and long term memory (STM, LTM).

Experiments

For evaluating the integrated model a Mixed Reality environment for the synthetic forager is being built which offers the possibility of making experiments in a controlled setting and then measure, analyze and compare behavioral data of the robot with a real rodent.

Figure 3. Left: Prototype system underlying the Synthetic Forager outdoor mobile platform. Right: The Khepera micro-robot (K-team,Lausanne) used in several of the DAC experiments measures about 55 by 30 mm and is equipped with a color CCD camera and active infra-red sensors that allow it to detect collisions and ambient light levels.

[1] Verschure, P.F.M.J., Voegtlin, T., Douglas, R.J., Environmentally mediated synergy between perception and behaviour in mobile robots, Nature (2003). 425:620-624 [2] Verschure, P.F.M.J. and Althaus, P., A real-world rational agent: Unifying old and new AI. Cognitive Science, 2003. 27: p. 561–590