Janne V. Kujala

Janne V. Kujala, Tuomas J. Lukka, Harri Holopainen

We consider the problem of sorting a densely cluttered pile of unknown objects using a robot. This yet unsolved problem is relevant in the robotic waste sorting business. By extending previous active learning approaches to grasping, we show a system that learns the task autonomously. Instead of predicting just whether a grasp succeeds, we predict the classes of the objects that end up being picked and thrown onto the target conveyor. Segmenting and identifying objects from the uncluttered target conveyor, as opposed to the working area, is easier due to the added structure since the thrown objects will be the only ones present. Instead of trying to segment or otherwise understand the cluttered working area in any way, we simply allow the controller to learn a mapping from an RGBD image in the neighborhood of the grasp to a predicted result---all segmentation etc. in the working area is implicit in the learned function. The grasp selection operates in two stages: The first stage is hardcoded and outputs a distribution of possible grasps that sometimes succeed. The second stage uses a purely learned criterion to choose the grasp to make from the proposal distribution created by the first stage. In an experiment, the system quickly learned to make good pickups and predict correctly, in advance, which class of object it was going to pick up and was able to sort the objects from a densely cluttered pile by color.

Janne V. Kujala, Tuomas J. Lukka, Harri Holopainen

We present a research picking prototype related to our company's industrial waste sorting application. The goal of the prototype is to be as autonomous as possible and it both calibrates itself and improves its picking with minimal human intervention. The system learns to pick objects better based on a feedback sensor in its gripper and uses machine learning to choosing the best proposal from a random sample produced by simple hard-coded geometric models. We show experimentally the system improving its picking autonomously by measuring the pick success rate as function of time. We also show how this system can pick a conveyor belt clean, depositing 70 out of 80 objects in a difficult to manipulate pile of novel objects into the correct chute. We discuss potential improvements and next steps in this direction.

Ehtibar N. Dzhafarov, Janne V. Kujala

We present general principles underlying analysis of the dependence of random variables (outputs) on deterministic conditions (inputs). Random outputs recorded under mutually exclusive input values are labeled by these values and considered stochastically unrelated, possessing no joint distribution. An input that does not directly influence an output creates a context for the latter. Any constraint imposed on the dependence of random outputs on inputs can be characterized by considering all possible couplings (joint distributions) imposed on stochastically unrelated outputs. The target application of these principles is a quantum mechanical system of entangled particles, with directions of spin measurements chosen for each particle being inputs and the spins recorded outputs. The sphere of applicability, however, spans systems across physical, biological, and behavioral sciences.