Bad rush to next Sea trials

We are near the end of a mad rush to get all the robotics equipment in our lab ready for the next sea trials. As always, no matter how much we think it will be a breeze, getting ready involves a lot of work and stress. Part of this is due to the constant pressure to push the bounds of what we can do and, once we get things stable, to add more complexity to the experiments being planned.

This year the part of the team from McGill is putting particular focus on three different classes of experiment: leaning-based robot guidance, optimal data collection given constraints on time, teleoperation and tele learning using Microsoft Robotics Studio, and enhanced controller design. See, I added a 4th experimental context just since the time I started this paragraph!

The learning-based guidance is based on work by my student Philippe Giguere where he uses a new learning rule to exploit the correlations is time that are present in most of the observations made in the real world. Using these temporal correlations, the process of learning about different classes of experience can be made easier. These classes of experience can be different terrain textures (as in our most recent RSS paper) where the robot learns how to adapt it's walking modes to the type of terrain it is on. For example, on slippery terrain a more "careful" low-speed gait is appropriate.