Professionally skilled excavator operators are familiar with the trick that allows them to send the bucket's contents beyond the reach of the boom. The neuro-network-equipped robo-excavator turned out to be a diligent learner, who has also targeted stone throwing further than the boom's reach. Next on the agenda is throwing bulk materials and improving the accuracy of the bucket's operation at various heights.
Researchers from Switzerland (ETH Zürich) reported on the training process of the robo-excavator's neural network for precise manipulation of the bucket's contents. The reinforcement learning-based neural network was trained to throw balls and stones at a specified target, which was beyond the boom's reach (up to 9.5 m with a boom reach of 7.5 m). Such operations will help robotics cope with a broader range of tasks while expending less energy on movements, and will also make its work safer.
The excavator performed grabs and throws with the bucket using two degrees of freedom, which was not rigidly attached to the boom. Throws were made both in a straight line, when only the boom was in operation, and with the cabin turning. In the latter case, accuracy was slightly lower, but in any case, the projectile deviated from the aiming point by no more than 30–40 cm.
Researchers trained the neural network based on a modified 12-wheeled Menzi Muck M545 excavator. Previously, they trained the excavator to perform a range of non-trivial operations, such as teaching it to build a stable wall from unprepared stone blocks. The excavator independently assessed the balance of the stones and constructed a sturdy stone enclosure. For precise autonomous work in the field, the excavator constructs a model of the surrounding space with the help of sensors installed on it, within which it performs specified operations.
Ընտանիք:
Ընտանիք: 3dnews.ru
