This was the last of our four week open meeting period. To make progress on the project, we need team members to commit for a few months, so we can have consistency from week to week. Although we were still missing some potential contributors from the FIRST Robotics Competition team, who are very busy with their FRC build season, it was clear that those who attended were dedicated to making this project a success.
We worked on three fronts during this meeting: Physical, software, and personality. The team members most interested in the sculpture, mechanism, and central nervous system worked to catalog KEN’s parts. To build a new robot, we need to know what to purchase based on what is already in KEN. A talented member of the team drew a detailed diagram of KEN’s neck mechanism.
Another group of team members interested in the software and intelligence experimented with enhancements to the face recognition software. In the Feb 3 meeting, the face recognition software was shown to be broken. Investigation was done during the week to identify the cause. Some speculative enhancements were removed and some parameters adjusted to get the system working better. Testing at the meeting showed that performance was improved, but still not good enough. Two brothers were confused (they look similar, but not identical). Two unrelated students were incorrectly linked as the same identity. One student was not successfully recognized. More work is needed to improve the algorithm further.
The final group worked on learning the mechanism to define KEN’s personality. Three computers were set up with KEN’s conversation software. The team members were shown how to log into the computers and edit the AIML files to define how KEN responds. Because of time constraints, we didn’t get very far with implementing new conversations. That will be an ongoing activity.
February 3, 2016, was the third of our four introductory meetings for the project. This meeting had three objectives: 1) Fix KEN, 2) Master face recognition, and 3) Design new conversations.
Since the ECPI demo the previous day had suffered from a breakdown in one of KEN’s Raspberry Pi computers, we assigned a team to investigate the root cause. They determined that the cause was a faulty micro-USB power plug. Great job, team!
Face recognition is one of the features of the robot which is really cool, but can be hard to demonstrate, because it is sensitive to a variety of conditions. The goal of this activity was to help the team become comfortable with this feature and learn how to demonstrate it. We set the goal of getting the robot to recognize everyone in the group. Since KEN was tied up with the repair team, KEN’s Uncle Ross (the LEGO predecessor prototype to KEN) was brought in to help.
We attempted face recognition with two team members before it was clear that something is wrong with the algorithm as currently implemented, because it was already getting confused. We talked briefly about how the algorithm matches faces, but concluded that some bug investigation is needed to find out why it wasn’t working correctly.
The final activity was to design new conversations for KEN. Everyone had experienced some of the awkward things KEN will say, so this was a chance to define how KEN should respond. The goal was to take these ideas and then show how they can be coded into the system. To support this, we had three systems set up to run KEN’s AI system. This way small teams could test out how KEN currently responds while thinking of improvements. Many good ideas were captured, and we got to see a little preview of how these ideas can be translated into code.
On Tuesday, February 2, 2016, KEN made an appearance at ECPI University in Raleigh, NC. Tony Gonzalez, an engineering student at ECPI, organized the event to promote IEEE student membership and the formation of a new IEEE student branch at ECPI. The event was well attended with 59 participants including students, professors, and organizers. The IEEE Humanoid Robot Project was invited to present KEN as an illustration of some of the activities that are happening through IEEE in Eastern North Carolina.
After an overview of IEEE and student membership benefits by Tony, Daniel demonstrated KEN. Maria Hunter also provided an overview of The Forge Initiative and our collaboration on the Humanoid Robot project. The meeting was scheduled for 30 minutes, and the formal presentation was completed in that time; however, audience questions and interaction continued for another 40 minutes, making it a very exciting and interactive event.
One downside to the event was that KEN’s look around behavior was not functioning. Investigation by the team at our weekly project meeting the following night determined that the micro-USB plug which powers one of the Raspberry Pi computers was not making a reliable connection and will need to be replaced.
At our Jan 27 meeting, we had about 30 people in attendance including an important visitor from the IEEE Foundation, which is providing the grant for our work this year. We were pleased to welcome Karen Galuchie, the Executive Director of the IEEE Foundation, who joined to observe the meeting. The participants rotated through three activities. The first activity involved setting up KEN and troubleshooting issues. The second activity was a discussion on the mechanical design of the neck mechanism and brainstorming on how to improve it. The third activity focused on how KEN interacts with people and designing approaches to improve the interaction.
During the troubleshooting activity, the team was asked to set up KEN without any guidance from the experts. Then, we tried to observe what went wrong and figure out how to correct it. Here’s a picture of some notes we captured on the tabletop whiteboard documenting some of the steps.
The steps to get KEN running are not usually this complicated, but in this case, the network cables were plugged in incorrectly (remember, the team was doing this without instructions), which caused the network to fail to start, which cascaded failures to other components. Most of the effort was to shut the systems down cleanly, and restart them after the network cable issue was corrected.