No abstract

The need for rapid and reliable robot deployment is on the rise. Imitation Learning (IL) has become popular for producing motion planning policies from a set of demonstrations. However, many methods in IL are not guaranteed to produce stable policies. The generated policy may not converge to the robot target, reducing reliability, and may collide with its environment, reducing the safety of the system. Stable Estimator of Dynamic Systems (SEDS) produces stable policies by constraining the Lyapunov stability criteria during learning, but the Lyapunov candidate function had to be manually selected. In this work, we propose a novel method for learning a Lyapunov function and a policy using a single neural network model. The method can be equipped with an obstacle avoidance module for convex object pairs to guarantee no collisions. We demonstrated our method is capable of finding policies in several simulation environments and transfer to a real-world scenario.

When a heat recovery ventilator is operating under winter conditions, the water vapor present in the exhaust airflow can lead to frost formation. The outside temperature at which frost formation occurs depends on many variables such as the heat exchanger plate temperature, the exhaust air humidity ratio, the exhaust airflow and the plate spacing. In this study, a new 2D frost formation model is proposed and applied to counterflow parallel plate heat exchangers. The method is based on a frost growth and densification model. The frost densification depends on the square root of the time and the ratio of supercooling and supersaturation degree. An energy balance equation for the heat conduction through the frost layer and the heat and mass transfer from the moist air to the frost layer is used as a convergence criterion on the frost surface temperature prediction. The proposed 2D model showed that the airflow from a 2.5 mm parallel plate spacing heat exchanger can be reduced as much as 33% over a 25 minutes period. While a larger plate spacing, such as a 4.0 mm spacing, is less prone to airflow reduction due to frost growth, less than 5% reduction over the same time period, the 2.5 mm spacing is still more efficient than the 4.0 mm spacing at the end of the 25 minutes period, with efficiencies of 77% and 55% respectively.

The need for rapid and reliable robot deployment is on the rise. Imitation learning (IL) has become popular for producing motion planning policies from a set of demonstrations. However, many methods in IL are not guaranteed to produce stable policies that can be used for motion planning.The generated policy may not converge to the robot target, reducing reliability, and may collide with its environment, reducing the safety of the system. Also, demonstration data is tedious to collect either through kinesthetic teaching or expert performing the task, both involving human labour. Stable estimator of dynamic systems (SEDS) produces stable policies by constraining the Lyapunov stability criteria during learning, but the Lyapunov candidate function has to be manually selected, which can result in unsolvable scenarios. In this work, we propose a novel method for learning a Lyapunov function and a policy using a single neural network model from automatically generating demonstration data in simulation. The method can be equipped with an obstacle avoidance module for convex object pairs to guarantee no collisions. We demonstrated our method is capable of finding policies in several simulation environments and transfers to real-world scenarios.