ZuFF

Here is a picture of the small-scale model augmented with a navigation map.

The map is created in the following way:

1. a standard mesh spans the whole table (in blue),
2. for each object, we define a protected zone where the forklift can not go, and the standard mesh is deactivated under this area,
3. navigation nodes are defined around each object, and dynamically linked to the standard mesh

In other words, we define a standard map which is then disturbed by the presence of objects. By using fixed nodes around the objects, we ensure that a path between two objects is available as soon as they are distant enough. This would not be the case if
we relied only on a regular mesh, because it would depend on the space between its nodes.

The first prototype of our environment has been built!
The tracking of objects with ARTag has been implemented, and after some complications the calibration between the camera and the beamer is done!

During the last two months, we’ve been refining the concept of our warehouse’s small scale model. We defined three roles (driver - designer - manager), that will be taken by apprentices in three different worlds (concrete - figurative - abstract), that should support the move from concrete to more abstract concepts. The main idea is to start with a very concrete representation of a warehouse, through physical objects, and progressively move towards abstract representations by removing them and showing underlying variables and concepts.

In the first world, apprentices take the role of a forklift driver and do exercises that let them learn concepts related to path optimization, the use of double-jeu, … The warehouse is represented in a very real fashion, with shelves, boxes and forklifts. Apprentices interact with the simulation by giving orders to forklifts, choosing the path they should follow. This situation is very close to their daily job, and should thus be a good way to introduce them to the use of the model.

In the second world, apprentices are now warehouse designers, and are responsible for the layout and organization of the warehouse. In this world, moving objects (boxes and forklifts) are not tangible anymore but have been replaced by a digital counterpart. This frees us from the time constraints implied by the use of robots, and allows us to make simulations that span over longer periods of time. In this world, apprentices discover the influence of the layout of shelves and the arrangement of products in the warehouse on their work efficiency.

In the third world, all the physical objects have now disappeared, and the display does not show a warehouse anymore but lists of products and charts. Apprentices have to work on information, accessing underlying variables like storage levels, last year sales, employees costs, … They take the role of a manager, and have to decide when to place orders, manage the amount of working forklifts and take high-level decisions about the warehouse organization.

We thus have a progression from a concrete representation, very close to the apprentices’ daily experience, towards a very abstract one, in which only information is displayed. The second world should make the transition easier, by adding some abstraction (simulated time, digital objects, …).

So what? Defining those worlds and roles give us the overall layout of the simulation and will help us a lot for organizing our ideas, create scenarios and plan the development phase that starts now.

After having the first prototype of the forklift robot that will be used in our small scale warehouse model, it’s now time to interest us to the tracking system. Our idea is to use ARTag, a well-known “Augmented Reality” system which is able to detect fiducial tags in a frame and augment it with drawings in real-time.

We are not planing to take advantage of the augmentation capabilities of ARTag, but just its ability to detect tags that will be put on robots and shelves. We hope that it will give us a good estimation of the different objects of the warehouse model, as well as their orientation.

For this purpose, we equipped Eliott, our forklift robot prototype, with a tag and we’ll try to track its position using a roof camera grabbing images from a table surrounded by walls where the robot will move.

The actual integration of this tracking tool will be achieved by a student, who will join our group for a semester project.

I attended a very interesting workshop at the CSCL Alpine Rendez-Vous, which was actually a sort of meeting for the three labs involved in the Leading House Technologies for Vocational Training that my project is part of. The members of the scientific board were present as well and it was a good opportunity to show them the work that has been done on the three different projects.

The five experts asked us a lot of questions and gave us a very interesting feedback about our current developments and a lot of good ideas for the future steps. Patrick and I presented the results of our observation round, the first developments of our small scale model of a warehouse and our recent discussions about possible research questions. The main one is to use our model to bring the apprentices from real situations, implemented in the physical model, to more abstract concepts, illustrated on a large screen.

Our main idea is to use our small scale model as a learning tool allowing apprentices to understand abstract logistics concepts by moving them from real physical situations (physical model) to more abstract ones (illustrated on a large screen). The experts raised a very interesting idea, that consists in using multiple representations to allow the apprentices to progressively move towards a more abstract model of logistics.

So what? A very interesting workshop, and a good feedback of the scientific board that will help me to refine the research question to be addressed by my project.

The first prototype of the forklift robots we’ll use in our warehouse’s small scale model came to electrical life last week!

It’s equipped with a compass and a radio frequency transceiver, which allow it to get its orientation and communicate with a remote computer. The two motors drive the front wheels, letting the forklift turn on itself, ant the fork is able to move up and down.

It’s still an early prototype that will undergo a lot of modifications, like the miniaturization of the electronics, the addition of hall effect sensors to compute the movements of the fork and a modification of the fork to allow it to reach the floor.

So what? This is the first step of the hardware design of our small scale warehouse model. We’ll now interest us in the design of the shelves, the tracking system and the software.

We are currently working on the design of the forklifts that will be used in our small-scaled warehouse. As stated in a previous post, the constraints are that it should be realistic and its size should correspond to a given scale.

First of all, we did a test with the e-puck, an educational robot developed at EPFL. Beside its advantages of being small and easy to program, this robot lacks a fork and has an unrealistic physical shape and behavior. We should thus develop a fork and modify its shape to make it more realistic.

We then moved toward toys, looking for some RC forklift that we could customize. We ordered two cheap ones, a realistically shaped Linde one and a Tamyia kit. The former doesn’t do the job as it is too slow and has a too large turning radius. The latter is more interesting, although looking quite different than a real one, as it seems quite easy to customize. The main problem is that the fork can’t very easily.

We also found a more expensive forklift (several hundreds of euros…), which is radio controlled and seems to fit better our constraints. We’ve ordered one of them and are waiting for its delivery. We didn’t buy a remote control (which comes separately) as we would like to interface it with a computer. More on that in the following weeks…

We are currently working on custom electronics, buying microcontroller boards, motors and sensors in order to modify existing toys.

So what? It seems that building a forklift for our model will take us quite a lot of time as no commercial product fits our needs. Read next posts for a description of ongoing developments.

Last month we had a meeting with the teachers of the vocational school of Yverdon that we met at the beginning of the project. The goal was to discuss with them our observations and the idea of building a small scaled model of a warehouse. Both of them are very interested in this project and thus ready to invest time in it.

We met again one week ago to discuss the possible scenarios that could be implemented in this physical simulation, as well as the different constraints that should be met. According to the teachers, it is very important for our model to be consistent with the reality. If not, apprentices wouldn’t take it seriously, considering it as meaningless.

The constraints are thus the following:

• every part of the warehouse should be at the same scale (e.g. 1/16),
• the physical behavior of the forklifts should be realistic,
• driving rules should be respected.

The teachers also think that apprentices shouldn’t have to drive the forklifts, but give them only high-level orders (e.g. take this pallet and move it to that position, following this path, …). This corresponds to what we had in mind, the goal being to teach apprentices management skills and not driving skills.

So what? These two meetings were very interesting and allowed us to confront our conclusions with people involved in vocational training. We’re glad to see that we got a good understanding of the apprentices’ problems and that our idea of a warehouse model was welcomed by the teachers.

A. Pentland and A. Liu. Modeling and prediction of human behavior. Neural Comput., 11(1):229–242, 1999.

This paper presents a model for modeling and prediction of human behavior, using a set of dynamic models (e.g. Kalman filter), coupled into a Markov chain. The authors call this type of model a Markov Dynamic Model (MDM).

Our proposal is to describe the small-scale structure of human behavior by a set of dynamic models (thus incorporating constraints such as smoothness and continuity) and the large-scale structure by coupling togheter these control states into a Markov chain.

MDM have been used to try to predict the behavior of automobile drivers. Subjects are driving in a simulator and from time to time a text command appears on the screen, asking them to turn left or right, stop, change lane, … The objective of the model is then to predict the action of the driver, based on the first few preparatory steps.

MDM have been compared to Bayesian classification. The former obtains a mean recognition accuracy of 95%, while the later does not perform better than chance. Here is the conclusion of the authors:

We believe that these results support the view that human actions are best described as a sequence of control steps rather than as a sequence of raw positions and velocities. In the case of driving, this means that it is the pattern of acceleration and heading that defines the action. There are lots of ways to manipulate the car’s controls to achieve a particular acceleration or heading, and consequently no simple pattern of hand-foot movement that defines a driving action.

So what ? Looking for a research question, and human behavior modeling may be an interesting subject, well-suited for the warehouse model we plan to build for logistics managers apprentices.

V. Aarkrog. Apprentices’ Transfer of Knowledge from School to Workplace in the VET Dual System: A study of a VET-Programme for Rescue Officers, in Working Knowledge in a Globalizing World - From Work to Learning, from Learning to Work, volume 3 of Studien zur Berufs- und Weiterbildung - Studies in Vocational and Continuing Education, pages 25–38. Peter Lang, 2006.

This paper studies the way Danish rescue officers apprentices transfer knowledge from school to workplace in two different activities: auto assistance and assistance in ambulance service.

Interviews about auto assistance have shown that apprentices are able to transfer knowledge from school and consider it useful. They don’t simply apply knowledge acquired at school but also use their practical experience. Experience is acquired rather quickly, and some critiques arose because what is learnt at school doesn’t always correspond to the reality.

Ambulance service is a much more difficult activity as it involves human interaction. In this case, apprentices also think that knowledge acquired at school is useful, but have this feeling for a longer time. This is due to the fact that apprentices can not simply apply rules or techniques learnt at school but have to adapt to each situation. The knowledge that is useful there is

“knowledge that will help them to estimate the situation”.

We thus have to different situations: auto assistance, where apprentices can apply more or less directly the knowledge acquired in school, and assistance in ambulance service, where apprentices need knowledge to interpret the situation, thus the concluding question of the paper:

“how may the school-based education teach the apprentice to use knowledge as a tool for interpretation ?”

So what ? Logistics managers apprentices’ work correspond to the first situation. Knowledge acquired at school can be applied directly to their job, but they learn it very quickly and their experience does not always correspond to what they learnt. On the other hand, the second situation that would correspond to the management part of their job, never occurs because they are not allowed to take part to this type of activity at the workplace.