10 Apr

CoLo has been redesigned for better compatibility and modularity so that it can be easier for user to modify the environment and adding new algorithm. So far, the majority of the new version of CoLo has been implemented. Later, we will add a more powerful analyzer to the environment! Also, we still...

10 Apr

The ML layer and the Gazebo simulation layers for RoCo were developed independently. I worked on the communication between these layers. Each of the layers is dependent on the other at certain stages. For example, for each simulation step, Gazebo needs new control parameters from the ML layer. S...

10 Apr

We can regard Coavariance Intersection as special case of the solution of the barycenter of several probability distributions, when the distance measure is KL divergence. By changing the distance measure, one can get different barycenter of those probability distributions. One of the most important...

07 Apr

I just installed Ubuntu on one of the lab computers. The Kinect detector now works on Linux and can detect depth information.

05 Apr

The Robot Compiler (RoCo) is a robot design tool to help the general population design and fabricate their own robots. This graphic explains the current abstraction model for representing and combining components. RoCo is modular and hierarchical, allowing designers to compose robots by adding sub...

19 Mar

As work on RoCo progresses it became clear that not all robots would necessarily have to be autonomous and that we would need to find a way to add user input in the design process. One of the previous versions of the paperbot had a html page loaded in the microcontroller, which would accessible from...

19 Mar

This week I spent a lot of time fixing stability bugs with the ORCAS system and gathering data with the reinforcement learning system. With a learning system, stability is extremely important as learning can takes days or even weeks which means that any small issue can completely bring down the whol...

12 Mar

For last week, I created flexible joint for furniture design to expand the design space. This flexible joint can allow two faces join togrther with certain angle while the finger joint only works well for 90 degree. But currently, our system has limited choices for one furniture design, which means...

12 Mar

This week the ORCAS team finished the communication layer between Gazebo and the reinforcement learning algorithm. With this, we were able to start training robots. We created a parameterized four legged robot and starting training a dense neural network (DNN) controller for it. Video In this sim...

12 Mar

This week, the ORCAS team set finished setting up our learning environment and began running the first tests. To do this, we first needed to create a parameterized robot design that we could use in our simulations. Our initial plan was to use a two wheeled robot as the robot for which we would opt...