Machine for dispensing a controlled amount of a cosmetic composition
The present application relates to a device for dispensing a determined weight of a cosmetic product into a receptacle from a reservoir, the device comprising at least one electromechanical member capable of moving an amount of cosmetic product from said reservoir installed in the device to a dispensing zone of said device where said amount of product may be transferred through a nozzle into the interior of the receptacle received in said dispensing zone, the device comprising a controller configured to deliver a control signal to the electromechanical member according to an error with respect to a setpoint corresponding to the determined weight of cosmetic product to be delivered, said error being determined on the basis of a weighing datum obtained by at least one weighing cell for the reservoir installed in the receiving zone, the device being characterized in that the controller is a reinforcement-learning controller.
1 . A device for dispensing a determined weight of a cosmetic product into a receptacle from a reservoir, the device comprising:
at least one electromechanical dispensing member capable of moving an amount of cosmetic product from said reservoir installed in the device to a dispensing zone of said device where said amount of product may be transferred through a nozzle into the interior of the receptacle received in said dispensing zone; and
a controller configured to deliver a control signal to the electromechanical dispensing member according to an error with respect to a setpoint corresponding to the determined weight of cosmetic product to be delivered, said error being determined on the basis of a weighing datum obtained by at least one weighing cell for the reservoir installed in the dispensing zone, wherein the controller is a reinforcement-learning controller.
2 . The device according to claim 1 , wherein the reinforcement-learning controller uses a regression algorithm according to a Gaussian process.
3 . The device according to claim 1 , wherein the reinforcement-learning controller uses a deep-reinforcement-learning algorithm.
4 . The device according to claim 1 , wherein the reinforcement-learning controller is based on an initial predictive model.
5 . The device according to claim 1 , wherein the reinforcement-learning controller operates without an initial predictive model.
6 . The device according to claim 1 , wherein the reinforcement-learning controller has been pre-trained prior to commercial use of the device.
7 . The device according to claim 1 , wherein the reinforcement-learning controller has not been pre-trained prior to commercial use of the device.
8 . The device according to claim 1 , wherein the reinforcement-learning controller is configured to perform an adaptation of control strategy following a plurality of dispensing operations, in particular after a determined number of dispensing operations and/or after a given period of time, on the basis of the operations performed during this period of time, together defining a dispensing profile.
9 . The device according to claim 1 , wherein the reinforcement-learning controller is configured to perform one or more adaptations of control strategy over a dispensing operation, in particular at regular intervals after a unitary control action.
10 . The device according to claim 1 , wherein the electromechanical dispensing member is a positive-displacement pump.
11 . The device according to claim 1 , wherein the electromechanical dispensing member is a syringe.
12 . The device according to claim 1 , wherein the device is configured to receive a plurality of reservoirs that are configured for functionally different cosmetic ingredients, said functionally different cosmetic products being capable of being combined so as to form a cosmetic product.
13 . The device according to claim 1 , wherein the reinforcement-learning controller implements a policy function that defines an action to be performed by the electromechanical dispensing member according to a state of the device during an operation of dispensing a cosmetic product, the state of the device comprising the weighing datum representing the weight of the cosmetic product dispensed into the receptacle and the error with respect to the setpoint.
14 . A method for controlling an electromechanical dispensing member with which a device according to claim 1 is equipped, the method comprising:
providing a setpoint representing a determined weight of product to be dispensed,
sending a control signal to the electromechanical dispensing member,
obtaining, from the weighing cell, the weight of product dispensed following the execution of the control command by the electromechanical dispensing member,
determining a difference between the weight of product dispensed following the execution of the control command and the setpoint,
modifying the control signal by means of a reinforcement-learning process with a view to minimizing the difference.
15 . The method according to claim 14 for controlling a device, wherein the control signal corresponds to a complete dispensing profile of the setpoint and is modified after at least one complete dispensing operation.
16 . The method according to claim 14 for controlling a device, wherein the control signal corresponds to a unitary incomplete dispensing action, the control signal being modified at least once over the complete dispensing process, the steps of sending the control signal to the electromechanical dispensing member, of obtaining and of determining the difference in dispensing and of adapting the control signal being repeated for as long as said obtained difference is greater than zero.