As a graduate level class CMPE215: Models of Robotic Manipulation used a project to demonstrate the student's comprehension of the course material. I teamed up with a classmate who showed me this video of Boston Dynamics' Handle robot. We couldn't find any technical information about the robot (in early 2019), so we decided to limit our analysis to the arm and estimate it's measurements for the project.
On the surface it seemed that analysing a five degree of freedom (DOF) arm wouldn't be too difficult, after all, we had just spent six weeks learning how to do that and practiced with more complicated examples. Two main snags caused us problems; the offset at the shoulder joint and the particular requirements of modelling the arm within Peter Corke's Robotic Toolbox for MatLab. The shoulder joint is really a third, very small, segment (rigid body) between joints, and once it was conceptualised that way it wasn't a problem. The problem I had with setting up the simulation is a good example of the difference between academic and real-life experience. My initial, and several iterations after that, homogeneous transforms and Denavit-Hartenberg (DH) parameters, where technically correct (I had other students go over them as well) but didn't work within the simulation. It took quite a bit of digging with the documentation, and some trial and error, to find the proper configuration expected by Peter Corke's Robotic Toolbox. With the kinks worked out it was quite fun, and very education, working in MatLab with Peter Corke's Robotic Toolbox, exploring the joint space of the arm, with the work leading to a real feeling for how this arm functions.