Brain-like decision-making and motion control system

A brain-like decision-making and motion control system for a robot, the brain-like decision-making and motion control system being implemented in embedded hardware and integrated into the robot, the brain-like decision-making and motion control system comprising an active decision-making module;

• an automatic decision-making module;
• an evaluation module;
• a memory module;
• a perceptual module;
• a compound control module;
• an input channel module;
• an output channel module;
• and a controlled object module, wherein the controlled object module comprises one or more executive elements,
• wherein the active decision-making module is configured to synthesize perceptual information from the perceptual module and memory information from the memory module so as to actively select a first target and a first operation in combination with decision rules and instruct the compound control module to perform the first operation, and to control the input channel module and the output channel module,
• wherein the automatic decision-making module is configured to automatically select a second target and a second operation to be performed according to an experience value, the perceptual information and the memory information when a computational load of the active decision-making module is heavy or when a prompt response is needed by the brain-like decision-making and motion control system, to instruct the compound control module to perform the second operation, and to assist the active decision-making module to make decisions,
• wherein the evaluation module is configured to evaluate a practical value of a performed target and an actual effect value of a performed operation through a feedback of an internal state or an external environment of the brain-like decision-making and motion control system, and to return a predictive error to the active decision-making module, the automatic decision-making module and/or the compound control module,
• wherein the memory module is configured to encode and record historical information of an internal state of the brain-like decision-making and motion control system and historical information of the external environment into the memory information, and to transmit the memory information to the active decision-making module and/or the automatic decision-making module to assist a decision process, and to transmit the memory information to the compound control module to affect a selection and an execution of action sequences,
• wherein the perceptual module is configured to process sensing information obtained from one or more sensors into the perceptual information, and to transmit the perceptual information to the active decision-making module, the automatic decision-making module, the evaluation module, the memory module, and the compound control module,
• wherein the compound control module comprises a plurality of basic control units, wherein each basic control unit comprises a plurality of neurons, wherein the plurality of neurons comprises at least one first neuron configured to accept input from the active decision-making module, wherein the plurality of neurons of each basic control unit comprises at least one second neuron configured to be connected with one or more neurons located in said each basic control unit to form unidirectional or reciprocal excitatory or inhibitory connections, wherein the plurality of neurons of each basic control unit comprises at least one third neuron configured to be connected with one or more neurons located in at least one basic control unit other than said each basic control unit to form the unidirectional or reciprocal excitatory or inhibitory connections, wherein the plurality of neurons comprises at least one fourth neuron configured to be connected to the output channel module or the evaluation module, wherein the connections are configured to learn, memorize, encode, control, anticipate or execute one or more action sequences through synaptic plasticity processes, and to assist the active decision-making module to perform among the action sequences,
• wherein the input channel module is configured to convert, relay, and amplify the perceptual information transmitted from the perceptual module, and output the perceptual information to the active decision-making module or the autonomous decision-making module and/or the compound control module respectively, wherein the input channel module is configured to accept a control signal input from the active decision-making module to adjust signal relay magnification,
• wherein the output channel module is configured to convert, relay, amplify an output signal of the compound control module, and to transmit the output signal to the controlled object module, wherein the output channel module is configured to accept a control signal input from the active decision-making module to adjust the signal relay magnification, and
• wherein the output signal is used by the output channel module to generate actuation control signals used by the controlled object module for the executive elements;
• wherein the active decision-making module comprises a target set, a target-operation-effect set, and a decision rule set,
• wherein the target set is configured to record one or more appeared targets that have appeared, and a respective comprehensive value and a respective urgency value of each appeared target, and to set the respective comprehensive value of each target that has not appeared as a first pre-set value to represent a curiosity value for each target that has not appeared,
• wherein the target-operation-effect set is configured to record one or more executable operations each is executable for a respective target, and to record a respective comprehensive effect value of each executable operation in this situation,
• wherein the decision rule set is configured to record the decision rules, wherein said decision rules comprise first decision rules defined by users and second decision rules newly learned by the brain-like decision-making and motion control system, wherein the decision rules are configured to specify how the brain-like decision-making and motion control system selects a potential selected target and a potential selected operation,
• wherein after a first potential selected target is selected by the active decision-making module, the active decision-making module is configured to transmit the first potential selected target and the respective comprehensive value of the first potential selected target to the evaluation module, and to select a first potential selected operation according to the first potential selected target from the target-operation-effect set,
• wherein the active decision-making module is configured to perform, if a second potential selected target and a second potential selected operation selected by the automatic decision-making module are transmitted from the automatic decision-making module to the active decision-making module, weighting on the first potential selected target and the first potential selected operation and the second potential selected target and the second potential selected operation to generate a final decision,
• wherein after the active decision-making module has decided to perform a decided operation according to the final decision, the active decision-making module is configured to transmit the executable operation and the comprehensive effect values of the executable operation to the evaluation module, and to transmit the executable operation in a form of an spiking sequence signal to the compound control module through an active-start-control pathway and an active-stop-control pathway,
• wherein when receiving a predictive error of the executable operation or the respective target of the executable operation transmitted from the evaluation module, the active decision-making module is configured to add the predictive error to a current comprehensive value of said corresponding target or a current comprehensive effect value of the executable operation to obtain an updated comprehensive value or an updated comprehensive effect value, and to update the updated comprehensive value and the updated comprehensive effect value into the target set and the target-operation-effect set, respectively.

A brain-like decision-making and motion control system for a robot, the brain-like decision-making and motion control system being implemented in embedded hardware and integrated into the robot, the brain-like decision-making and motion control system comprising an active decision-making module;

• an automatic decision-making module;
• an evaluation module;
• a memory module;
• a perceptual module;
• a compound control module;
• an input channel module;
• an output channel module;
• and a controlled object module, wherein the controlled object module comprises one or more executive elements,
• wherein the active decision-making module is configured to synthesize perceptual information from the perceptual module and memory information from the memory module so as to actively select a first target and a first operation in combination with decision rules and instruct the compound control module to perform the first operation, and to control the input channel module and the output channel module,
• wherein the automatic decision-making module is configured to automatically select a second target and a second operation to be performed according to an experience value, the perceptual information and the memory information when a computational load of the active decision-making module is heavy or when a prompt response is needed by the brain-like decision-making and motion control system, to instruct the compound control module to perform the second operation, and to assist the active decision-making module to make decisions,
• wherein the evaluation module is configured to evaluate a practical value of a performed target and an actual effect value of a performed operation through a feedback of an internal state or an external environment of the brain-like decision-making and motion control system, and to return a predictive error to the active decision-making module, the automatic decision-making module and/or the compound control module,
• wherein the memory module is configured to encode and record historical information of an internal state of the brain-like decision-making and motion control system and historical information of the external environment into the memory information, and to transmit the memory information to the active decision-making module and/or the automatic decision-making module to assist a decision process, and to transmit the memory information to the compound control module to affect a selection and an execution of action sequences,
• wherein the perceptual module is configured to process sensing information obtained from one or more sensors into the perceptual information, and to transmit the perceptual information to the active decision-making module, the automatic decision-making module, the evaluation module, the memory module, and the compound control module,
• wherein the compound control module comprises a plurality of basic control units, wherein each basic control unit comprises a plurality of neurons, wherein the plurality of neurons comprises at least one first neuron configured to accept input from the active decision-making module, wherein the plurality of neurons of each basic control unit comprises at least one second neuron configured to be connected with one or more neurons located in said each basic control unit to form unidirectional or reciprocal excitatory or inhibitory connections, wherein the plurality of neurons of each basic control unit comprises at least one third neuron configured to be connected with one or more neurons located in at least one basic control unit other than said each basic control unit to form the unidirectional or reciprocal excitatory or inhibitory connections, wherein the plurality of neurons comprises at least one fourth neuron configured to be connected to the output channel module or the evaluation module, wherein the connections are configured to learn, memorize, encode, control, anticipate or execute one or more action sequences through synaptic plasticity processes, and to assist the active decision-making module to perform among the action sequences,
• wherein the input channel module is configured to convert, relay, and amplify the perceptual information transmitted from the perceptual module, and output the perceptual information to the active decision-making module or the autonomous decision-making module and/or the compound control module respectively, wherein the input channel module is configured to accept a control signal input from the active decision-making module to adjust signal relay magnification,
• wherein the output channel module is configured to convert, relay, amplify an output signal of the compound control module, and to transmit the output signal to the controlled object module, wherein the output channel module is configured to accept a control signal input from the active decision-making module to adjust the signal relay magnification, and
• wherein the output signal is used by the output channel module to generate actuation control signals used by the controlled object module for the executive elements;
• wherein each basic control unit comprises an active control layer;
• a motion encoding layer;
• a disinhibition layer;
• and a relay layer,
• wherein the active control layer comprises an initiating neuron (IN) and a stopping neuron (SN),
• wherein the motion encoding layer comprises an initiating action encoding neuron (IAEN) and a stopping action encoding neuron (SAEN),
• wherein the disinhibition layer comprises an initiating disinhibitory neuron (IDIN) and a stopping disinhibitory neuron (SDIN),
• wherein the relay layer comprises an initiating relay neuron (IRN) and a stopping relay neuron (SRN),
• wherein in the active control layer, the IN is configured to receive an active start signal from the active decision-making module and an excitatory connection of the IRN, the SN is configured to receive an active stop signal from the active decision-making module, the IN and the SN are configured to respectively synthesize and encode input signals into spiking signals and output the spiking signals to the IAEN and the SAEN in the motion encoding layer;
• wherein in the motion encoding layer, the IAEN is configured to receive excitatory input signals from the IN, and the SAEN is configured to receive the excitatory input signals from the SN, wherein the IAEN and the SAEN are configured to receive the input signals from the memory module, the perceptual module and the evaluation module, wherein the IAEN and the SAEN are configured to respectively synthesize the input signals to generate the spiking signals so as to modulate pulse discharge frequency and phase of the IDIN and the SDIN in the disinhibition layer,
• wherein one or more reciprocal inhibitory connections are between the IAEN and the SAEN in a same basic control unit, wherein relative action intensity and relative duration of the meta-action corresponding to said same basic control unit are configured to be encoded by synaptic weights,
• wherein the reciprocal inhibitory connections are between the IAENs in multiple basic control units, wherein an execution sequence of a respective meta-action in the action sequences is configured to be encoded by the synaptic weights,
• wherein the inhibitory connections are between the IAENs and the SAENs in different basic control units, wherein a relative start time and a relative stop time of the respective meta-action in the action sequences are configured to be encoded by the synaptic weights,
• wherein in the disinhibition layer, the IDIN is configured to receive inhibitory input signals from the IAEN and the SDIN, and the excitatory input signals from the SRN, and to synthesize all the input signals to generate the spiking signals, and to modulate the pulse discharge frequency and phase of the IRN through the inhibitory connections, wherein the SDIN is configured to receive the inhibitory input signals from the SAEN and the excitatory input signals from the SRN, and to synthesize all the input signals to generate the spiking signals, and to modulate the pulse discharge frequency and phase of the SRN through the inhibitory connections,
• wherein in the relay layer, the IRN is configured to receive the inhibitory input signals from the IDIN, and to form a positive feedback closed loop so as to promote motion execution through the excitatory connection projected to the IN, wherein the SRN is configured to receive the excitatory input signals from the SN and the inhibitory input signals from the SDIN, to synthesize all the input signals to generate the spiking signals, and to project the spiking signals to the IDIN through the excitatory connection to form a pathway for inhibiting execution of actions,
• wherein the IN, the SN, the IAEN, the SAEN, the IDIN, the SDIN, the IRN and the SRN are served as outputs of the basic control units.