The tangential strategy

In this example video, Ernest 8.2 found a strategy that we named the tangential strategy.

The tangential strategy consists of approaching the blue square in a straight line as opposed to a diagonal line (the diagonal strategy in the previous example). The trick with the tangential strategy is that Ernest cannot know when he should turn toward the blue square until he passed it. The tangential strategy thus consists of moving on a straight line until the blue square disappears from the visual field, then returning one step backward, and then turning toward the blue square.

The emergence of a specific strategy occurs during Ernest's youth while he his babbling relatively randomly, in parallel to the emergence of goals. See the details in the next post. When Ernest has organized behavioral patterns that proved both satisfying and robust, he adopts them and stick to them as long as they work.

These results demonstrate that:
a) Ernest does not encode strategies nor task procedures defined by the programmer, as opposed to traditional cognitive modeling.
b) Ernest instances are capable of "individuating" themselves through their experience, i.e., acquiring their own cognitive individuality that was not encoded in their "genes". This accounts for the role of individual experience in cognitive development.
c) Ernest's goals emerge from his low-level drives. Eating blue squares appears to the observer as becoming the goal of Ernest' life while no representation of such goal was encoded into Ernest. Indeed, Ernest was given a high incentive to step on blue squares but this incentive was not different in nature from other primitive drives. Ernest's goals were not pre-encoded as they are, for example, in the goal buffer of the ACT-R architecture.

Ernest 8.2 can find his food

Ernest 8.2 is a horseshoe crab. Horseshoe crabs are archaic arthropods whose visual system has been extensively studied. From these studies, we pulled several principles that guided the development of Ernest's distal sensory system:

- Small matrix resolution: the horseshoe crab's most elaborated eyes (two compound eyes among the 10 eyes that horseshoe crabs possess) have a resolution of roughly 40*25 pixels.
- Fixed eyes: eyes are fixed to the animal's body. The animal has to rotate its full body to move its visual field.
- Sensibility to movement: the signal sent to the brain does not reflect static shape recognition but rather reflects changes in the visual field.
- Visio-spatial behavioral proclivity: male horseshoe crabs move toward females when they see them with their compound eyes whereas females move away from other females.

As noted earlier, Ernest's "eyes" have only one pixel — pixel sensible to the distance to the blue square in a 90° visual field (assuming there is only one blue square).

Each eye produces a signal that represents the change in the corresponding visual field during the last interaction cycle:
- Appear: a blue square appeared in the visual field.
- Closer: more blue in the visual field, meaning the blue square is approaching.
- Arrived: the blue square occupies the entire visual field, meaning Ernest is stepping on the blue square and can eat it.
- Disappeared: the blue square disappeared from the visual field.

As opposed to previous versions, Ernest 8.2 has no antenna and has only three possible primitive behaviors:
- [move forward, succeed, 0] Ernest is indifferent of moving forward.
- [move forward, fail, -8] Ernest hates bumping walls.
- [turn left or right, succeed, 0] Ernest is indifferent of turning toward an adjacent empty square.
- [turn left or right, fail, -5] Ernest dislikes turning toward an adjacent wall.

To generate a visio-spatial behavioral proclivity, Ernest's sequential learning mechanism receives an additional inborn intrinsic satisfaction when an eye returns a signal:
- [Appear, 15] Ernest loves blue squares appearing in an eye's visual field.
- [Closer, 10] Ernest enjoys blue squares getting closer.
- [Arrived, 30] Ernest is crazy about stepping on a blue square (and eating it in the process).
- [Disappear, -15] Ernest hates blue squares disappearing from an eye's visual field.

At the beginning, the video shows Ernest learning to coordinate his actions with his sensory input. As before, he needs to learn to generate expectations associated with actions (e.g., turning schemas may shift the blue square from one eye to the other, etc.). He also needs to learn sequences of behavior (or "strategies") to reach the blues square. In this example run, he learned a strategy consisting of following a diagonal, and subsequently a straight line. Other strategies are possible that we will report next.

Ernest 8.1 can eat

Ernest can now perform different actions with different body parts simultaneously. Namely, he can eat this delicious blue substance with his virtual mouth while enacting his usual wandering behavior with his antennas and virtual legs.

To obtain this result, Ernest was provided with a modular internal structure. Ernest now has an iconic module that implements his first skills to exploit his distal sensory system, and a "homeostatic" module that controls his homeostatic regulation behavior. Ernest 8.1, yet, only has one single homeostatic behavior: eating the blue substance. Each module has its own enaction canal, making simultaneity across modules possible.

The homeostatic primitive schema was added to the other primitive schemas listed previously:
- [Eat, succeed, 100] Ernest is crazy about eating.
and one iconic pattern was predefined:
- [0,0] Blue_icon (each pixel represents the distance to the blue square in the corresponding sensory field).

Ernest now also supports inborn composite schemas. In this experiment, these are:
- [Touch ahead empty, Move forward, 3]
- [Turn left toward empty, Move forward, 3]
- [Turn right toward empty, Move forward, 3]
- [Blue_icon, Eat, 3]

The three first inborn composite schemas are just provided to accelerate the initial learning. The fourth is provided to make Ernest eat when he senses that he is on a blue square. These four composite schemas were preset with an initial weight of 3.

The sensory system being linked to the homeostatic system through the [Blue_icon, Eat] composite schema makes Ernest 8.1's distal sensory system somehow resemble smell (in addition to vision) because the sensory system detecting the null distance to food triggers eating.

These developments were inspired by Joanna Bryson's paper Structuring Intelligence: The Role of Hierarchy, Modularity and Learning in Generating Intelligent Behaviour. Joanna's discussion on the agent's modular architecture inspired our implementation of Ernest's modules. Also, we follow her argument that sometimes it is ok to pre-encode the desired behavior into the agent.

Now the question is how to implement the effects that the satisfaction gained through eating should have on Ernest to make him exploit his distal sensors in the search for food.

Ernest 8.0 has rudimentary distal perception

Ernest 8.0's distal sensory system consists of two pixels that react to blue squares. Each pixel reflects 90° of Ernest's surrounding environment. Ernest's left-side pixel reflects the front-left 90° quadrant, and Ernest's right-side pixel reflects the front-right 90° quadrant.

We can think of such a sensory system as an initial visual system. The pixel's value reflects the amount of blue in the corresponding visual field. If there is only one blue square, the pixel's value reflects the blue square's distance from Ernest. Green squares are opaque, meaning that blue squares behind green squares are not detected.

In the beginning of this example video, Ernest is trained the same way as in previous experiments. Then, the environment is opened and a blue square is added. We can see the two pixels on Ernest's head that reflect the detected blue. The closer the blue square, the more vivid the corresponding blue pixel.

Ernest 8.0 does not yet use this sensory system to inform his behavior. Making bottom-up intrinsic motivation and distal sense work together will be one of our next challenges.

Java Ernest 7.2 in Vacuum

We put Ernest 7.2 back to the Vacuum environment to provide an example implementation of the Ernest Java class.

As we can see in this example video, the Java version is much faster than the Soar version. In fact, we had to slow the Java version down with timers to be able to see something.

This experiment uses the following settings:
- [move forward, succeed, 5] Ernest enjoys moving forward.
- [move forward, fail, -8] Ernest hates bumping walls.
- [turn left or right, succeed, 0] Ernest is indifferent of turning toward an empty square.
- [turn left or right, fail, -5] Ernest dislikes turning toward a wall.
- [touch, succeed, -1] Ernest slightly dislikes touching walls.
- [touch, fail, 0] Ernest is indifferent of touching empty squares.

Java Ernest available

A Java version of Ernest is now available at http://code.google.com/p/e-ernest/. Many thanks to Mark Cohen for his contribution.

This version is fully available in read only. People willing to contribute to the project are very welcome. Please just contact me for full access.

The legend of Ernest 7.2

We are now working on moving Ernest to a 3D-rendering world, using Blender.

This preliminary video is intended as a legend to help interpret the Ernest 7.2's video shown in this older post.

The video shows Ernest's six primitive sensorymotor schemas, and illustrates the feedback that Ernest receives when enacting them in different situations:

- Touch ahead, to the right, and to the left. Touched squares flash yellow.
- Turn to the right or to the left.
- Move forward. When Ernest bumps into a wall, the wall flashes red.

Again, Ernest 7.2 has no visual perception of his environment. His only internal representation of his situation is that made through enacting sensorymotor schemas.

An Algorithm for Self-Motivated Hierarchical Sequence Learning

This six-page paper offers an overview of Ernest 7.2 algorithm and implementation:

Olivier L. Georgeon, Jonathan H. Morgan, Frank E. Ritter, (2010). An Algorithm for Self-Motivated Hierarchical Sequence Learning. Proceedings of the International Conference on Cognitive Modeling. Philadelphia, PA. ICCM-164, pp 73-78.