A biomimetic three-domain hierarchical decoupling and dynamic power interlocking robot intelligent control system

Abstract

The application belongs to the technical field of robots, and relates to a bionic three-domain hierarchical decoupling and dynamic power consumption interlocking robot intelligent control system. Three physical separated computing domains, namely a logical cognitive domain unit, a perception interaction domain unit and a motion control domain unit, are dynamically allocated with electric energy. The logical cognitive domain unit is responsible for global cognitive reasoning and task planning, the perception interaction domain unit is responsible for environment perception and data preprocessing, and the motion control domain unit is responsible for distributed motion execution and local closed-loop control. The three-domain physical layering of perception, decision and motion control is achieved, and a collaborative control mechanism of three-domain independent power supply regulation and cross-domain power consumption interlocking is established.

Description

[0001] This invention relates to the field of robotics, and more particularly to a biomimetic three-domain hierarchical decoupling and dynamic power consumption interlocking intelligent control system for robots. Background Technology

[0002] The closest existing technology in this field is a decision-motion control dual-layer centralized control system, which does not achieve physical layering of the three domains of perception, decision-making, and motion, nor does it have an independent power supply and dynamic power consumption interlocking mechanism.

[0003] Existing robot control systems generally adopt a two-layer centralized architecture of "decision layer - motion control layer", which couples functions such as environmental perception, logical reasoning, and motion planning onto limited computing resources, resulting in the following inherent defects:

[0004] 1. The computing resources for perception processing, logical reasoning, and motion control are mixed, resulting in high functional coupling and making modular upgrades and iterations difficult;

[0005] 2. Deep inference tasks and high-dynamic motion tasks are prone to being under high load at the same time, resulting in concentrated power consumption and heat dissipation pressure on the whole machine, which can easily cause processor throttling, system lag, and affect control stability;

[0006] 3. It cannot simulate the domain-based collaborative working mode of biological nervous systems; high-load reasoning and high-dynamic motion cannot be executed at off-peak times, and tasks are mutually restrictive.

[0007] 4. Lack of zoned power supply and dynamic power consumption scheduling mechanism results in low energy utilization efficiency and limited overall battery life;

[0008] 5. The joint control unit lacks local fault isolation and functional compensation mechanisms, making it prone to machine shutdown due to single-point failures, resulting in insufficient reliability.

[0009] There is a prevalent technical bias in this field of 'computing redundancy preference,' which advocates keeping all computing units online as much as possible to avoid task response delays. This bias has resulted in problems such as concentrated power consumption, high heat dissipation pressure, and system frequency throttling remaining unresolved for a long time.

[0010] Existing technologies have not achieved a three-domain physical layering of perception, decision-making, and motion control into logical cognition domain, perception interaction domain, and motion regulation domain, nor have they established a collaborative control mechanism that enables independent regulation of power supply in the three domains and interlocking of power consumption across domains. Summary of the Invention

[0011] The technical problem to be solved by this invention is: in order to realize the physical layering of perception, decision-making and motion control into three domains—logical cognition domain, perception interaction domain and motion regulation domain—and establish a collaborative control mechanism with independent power supply regulation in the three domains and cross-domain power consumption interlocking, this invention provides a biomimetic three-domain layered decoupling and dynamic power consumption interlocking robot intelligent control system.

[0012] The technical solution adopted by this invention to solve its technical problem is: a biomimetic three-domain hierarchical decoupling and dynamic power consumption interlocking robot intelligent control system, comprising:

[0013] The system consists of three physically separate computational domains: a logical cognition domain unit, a perceptual interaction domain unit, and a motion control domain unit, with energy dynamically allocated among the three domains.

[0014] The logical cognition domain unit is responsible for global cognitive reasoning and task planning, the perception and interaction domain unit is responsible for environmental perception and data preprocessing, and the motion control domain unit is responsible for distributed motion execution and local closed-loop control. The perception and interaction domain unit is the core of the overall control of the machine, coordinating the collaborative work of the logical cognition domain unit and the motion control domain unit.

[0015] It also includes: data interaction between the three computing domains via communication methods.

[0016] The perception and interaction domain unit actively and dynamically allocates power supply and computing resources of each domain according to the current overall machine operation requirements and the fixed total energy resources of the overall machine, so as to realize the coordinated scheduling and interlock protection of cross-domain power consumption.

[0017] This invention is an architecture-level invention. The features such as the three-domain layering, the QuickThink knowledge base, and the power consumption interlock are all definitions at the system architecture level, and do not limit the specific communication protocol, similarity matching algorithm, experience data storage format, or specific implementation method of power supply scheduling.

[0018] As a preferred embodiment of the present invention, the logical cognitive domain unit includes a high-performance main processor, a non-volatile memory module, and a global task scheduling engine; the global task scheduling engine is used to prioritize all tasks and adjust the task execution timing according to the feedback from the cross-domain dynamic power consumption interlock scheduling module.

[0019] As a preferred embodiment of the present invention, the perception interaction domain unit includes a multi-sensor fusion unit, an edge computing acceleration chip, and a real-time preprocessing module; the real-time preprocessing module is used to filter, extract features, and compress data from the raw sensor data, and only send structured data to the logical cognition domain unit.

[0020] As a preferred embodiment of the present invention, the motion control domain unit adopts a distributed architecture, and each joint drive subunit has independent drive control and state feedback capabilities, which can realize local closed-loop control of a single joint.

[0021] As a preferred embodiment of the present invention, energy is uniformly allocated, and each independent power supply module in the computing domain includes independent energy monitoring, power conversion and power management; the power supply links of the three power supply modules are physically isolated, and power regulation and on / off control can be performed independently.

[0022] As a preferred embodiment of the present invention, the cross-domain dynamic power consumption interlock scheduling module includes a power consumption acquisition module, a threshold dynamic adjustment module, a power consumption interlock execution module, and a safety protection module; the power consumption acquisition module collects the power supply parameters (including power supply, voltage, etc.) and temperature data of each domain in real time; the threshold dynamic adjustment module dynamically adjusts the power consumption limit of each domain according to the current task type and energy reserve of the machine.

[0023] As a preferred embodiment of the present invention, the motion control domain unit has a single-point fault isolation and functional compensation mechanism; when a certain joint drive subunit fails, its independent microcontroller automatically cuts off its own power supply, the distributed motion master controller immediately adjusts the motion trajectory, and the adjacent joint drive subunit performs functional compensation.

[0024] As a preferred embodiment of the present invention, the global task scheduling engine incorporates a dynamic task scheduling algorithm based on power consumption interlocking, which performs time-sliced ​​scheduling of deep inference tasks and high-dynamic motion tasks to avoid both types of tasks being under high load at the same time.

[0025] As a preferred embodiment of the present invention, the safety protection module integrates overcurrent, overvoltage, overheat and short circuit protection functions; when any domain experiences abnormal power consumption or hardware failure, the power supply to the corresponding computing domain is immediately cut off, and a graded safety shutdown sequence is triggered, while the fault information is reported to the logical cognition domain unit.

[0026] The beneficial effects of this invention are:

[0027] 1. Adopting a biomimetic three-domain hierarchical architecture design, the robot's intelligent behavior logic is more in line with the biomimetic hierarchical collaborative characteristics;

[0028] 2. By using a three-domain computing power layered decoupling and dynamic power consumption interlocking mechanism, idle computing power is managed for low-power sleep, which greatly reduces the overall power consumption of the machine and significantly improves the terminal's battery life.

[0029] 3. Adopting a three-domain independent heat dissipation and power consumption staggered operation design, the overall heat load is significantly reduced, avoiding overheating, frequency reduction, and crashes caused by high heat generation during simultaneous logic reasoning and physical movement.

[0030] 4. The hardware and algorithms of each domain are modularly and independently laid out, which facilitates technology iteration, maintenance and repair, and large-scale mass production.

[0031] 5. It has the ability to autonomously isolate and compensate for joint failures, enabling local fault-tolerant operation and ensuring continuous and uninterrupted operation of the entire machine;

[0032] 6. Through dynamic and intelligent allocation of resources in three domains, it can flexibly adapt to robot models of different configuration levels, different product positioning, and any form, including but not limited to legged, wheeled, tracked, humanoid, bionic animal, industrial robotic arms, composite robots, and other bionic or non-bionic configurations.

[0033] 7. It breaks the technical bias of "computing power redundancy preference" in this field and achieves Pareto improvement in system-level power consumption by actively reducing the power supply and computing power of unnecessary domains.

[0034] 8. Significantly improved real-time interaction capabilities: The QuickThink knowledge base supports extremely low latency response to known scenarios, greatly reducing the robot's execution latency for daily and repetitive tasks, and the smoothness of human-computer interaction is significantly better than that of a pure inference-driven architecture; at the same time, it further reduces the frequency of wake-up in the logical cognition domain, and the average power consumption of the whole machine is further optimized on the basis of the existing interlocking mechanism. Attached Figure Description

[0035] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0036] Figure 1 This is a diagram of the overall biomimetic three-domain architecture of the present invention;

[0037] Figure 2 This is a flowchart of the biomimetic three-domain dynamic power consumption interlocking process of the present invention;

[0038] Figure 3 This is a schematic diagram of the three-domain independent heat dissipation of the present invention. Detailed Implementation

[0039] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0040] Example 1:

[0041] like Figures 1 to 3 As shown, a biomimetic three-domain hierarchical decoupling and dynamic power consumption interlocking robot intelligent control system includes three physically separated computing domains: a logic cognition domain unit, a perception and interaction domain unit, and a motion control domain unit. Each computing domain dynamically allocates electrical energy.

[0042] The three-domain hierarchical architecture of this invention directly corresponds to the functional division of labor in the biological nervous system:

[0043] Perceptual interaction domain (including the QuickThink knowledge base) Brain related functional areas Perceive the outside world, draw upon existing experience, and react quickly ("practice makes perfect") Motion control domain Cerebellum region Maintaining balance, executing gait, and instinctive movements ("body memory") Logical cognitive domain Prefrontal cortex region Complex reasoning, deep thinking, and solving unfamiliar problems ("deliberate consideration")

[0044] The logical cognition domain unit is responsible for global cognitive reasoning and task planning, the perception and interaction domain unit is responsible for environmental perception and data preprocessing, and the motion control domain unit is responsible for distributed motion execution and local closed-loop control. The perception and interaction domain unit is the core of the whole machine control, coordinating the collaborative work of the logical cognition domain unit and the motion control domain unit.

[0045] The perception and interaction domain includes the main control module, the perception module, and the storage module. The perception and interaction domain is equipped with a scalable, high-capacity local storage module to store scene knowledge base, visual features, and environmental memory.

[0046] The perception and interaction domain is responsible for environmental perception, scene recognition, road condition judgment, and scene decision-making using binocular vision, radar, and all-body sensors.

[0047] The main control module is the sole master control unit for the entire machine, coordinating communication, task priority, and mode switching among all modules. The main control module dynamically configures motion parameters in the motion control domain and can directly issue single-limb control commands. It can also issue commands to the motion control domain to temporarily disable the flexible balance protection mechanism to execute special actions. In this embodiment, the motion parameters include, but are not limited to, one or more of parameters such as stride length, stride distance, stride frequency, and joint angle.

[0048] The perception center module is responsible for environmental perception and scene recognition, and the main control center module of the whole machine makes scene decisions based on the perception results;

[0049] The storage module is a scalable, high-capacity local storage module, used to store scene knowledge bases, visual features, and environmental memories;

[0050] The perception and interaction domain unit is responsible for the functions of power distribution, power management, dynamic power consumption scheduling and interlock management of the whole machine.

[0051] Meanwhile, the perception and interaction domain unit is also equipped with the Quick Thinking Knowledge Base subsystem, which enables autonomous shallow thinking and rapid decision-making, freeing it from dependence on deep reasoning in the logical cognition domain.

[0052] After the perception and interaction domain acquires environmental features, the QuickThink knowledge base subsystem is responsible for rapid matching and evaluation; for successfully matched scenarios, the system directly generates control commands and sends them to the motion control domain for execution, achieving instantaneous response with extremely low latency;

[0053] When the QuickThink knowledge base subsystem determines that the current scene exceeds its shallow reasoning ability, the perception interaction domain unit calls the logical cognition domain unit to perform deep reasoning.

[0054] The core capability of the QuickThink Knowledge Base subsystem lies in its built-in lightweight reasoning model and experience data. It can perform independent, shallow thinking and autonomous decision-making (such as obstacle avoidance, following, trajectory planning, and standard interaction) for most everyday, known, and structured scenarios without requiring deep reasoning from the logical cognitive domain. Its operating mode is as follows: after the perception and interaction domain acquires environmental features, the QuickThink Knowledge Base is responsible for rapid matching and evaluation. For successfully matched scenarios, the system directly generates control commands and issues them to the motion control domain for execution, achieving extremely low-latency instantaneous response. The knowledge source is pre-built experience: high-frequency basic behavioral strategies built in at the factory.

[0055] Deep reasoning results are stored in the database: When the logical cognitive domain has to be called to conduct deep thinking, once the reasoning conclusion it outputs is verified to be correct and valid, the perceptual interaction domain will distill the conclusion into knowledge and solidify it into the Quick Thinking knowledge base, transforming it into its own shallow experience for direct use in the future.

[0056] Meanwhile, the QuickThink knowledge base subsystem can continuously evolve. Through a closed-loop mechanism of "thinking when encountering new things (deep reasoning), learning when thinking (entering the database), and using what has been learned (shallow decision-making)," the robot becomes more and more intelligent with use, the reliance on deep reasoning in daily operation continues to decrease, and the average power consumption and response latency of the system are optimized in sync.

[0057] The QuickThink Knowledge Base-driven dual-path decision architecture of "autonomous shallow thinking + on-demand deep reasoning" builds a lightweight experience storage and fast matching subsystem (QuickThink Knowledge Base) within the perception and interaction domain, enabling instantaneous response to known scenarios and on-demand deep reasoning for unknown scenarios, further reducing the number of invalid wake-ups in the logical cognition domain, and coordinating with the power consumption interlock mechanism to optimize the overall energy consumption.

[0058] The core features of this mechanism include:

[0059] Shallow thinking path: In everyday known scenarios, decisions are made directly by the quick thinking knowledge base, with instantaneous response (far lower than deep reasoning path), and the logical cognition domain remains in low-power sleep mode;

[0060] Deep reasoning path: Deep reasoning is only invoked when the QuickThink knowledge base determines that the scenario exceeds the shallow level of capabilities;

[0061] Verification and storage loop: After the deep reasoning results are verified to be correct, they are distilled, compressed, and stored in the QuickThink knowledge base to achieve continuous evolution of "thinking when encountering new things, learning when thinking, and applying what has been learned".

[0062] like Figures 1 to 3 As shown, the logical cognition domain includes a logical algorithm subunit and a deep learning reasoning dedicated unit. The logical cognition domain unit is used for mathematical algorithms, logical deduction, deep learning of complex tasks, abstract theoretical calculation, and high-order strategy reasoning.

[0063] The logical cognition domain unit accepts the calculation and reasoning requirements of the perception interaction domain unit. After the calculation is completed, it sends the reasoning result back to the perception interaction domain unit, which then makes the final decision and executes the calculation.

[0064] The logical cognitive domain only receives instructions and computational requests from the perceptual interaction domain, and only sends results back to the perceptual interaction domain; it has no authority to issue instructions independently. There is no direct data link between the logical cognitive domain and the motion control domain; all data interactions must be relayed through the perceptual interaction domain. The logical cognitive domain refuses to execute any instructions from the motion control domain.

[0065] The motion control domain unit adopts a distributed architecture, and each joint drive subunit has independent drive control and state feedback capabilities, which can realize local closed-loop control of a single joint.

[0066] The motion control domain uses a dedicated real-time motion control processor as the core processing unit. It does not require the participation of artificial intelligence models and is entirely based on non-intelligent deterministic motion algorithms, resulting in fast response speed and stable and reliable operation.

[0067] This domain solidifies basic instinctive movements and advanced motor parameters: standing, balancing, squatting, lying down, rolling, walking, running, and independent control of single joints of the limbs;

[0068] Motion parameters, such as stride length, stride distance, cadence, leg lift height, joint angle, and movement speed, adopt an open and configurable mode and are dynamically adjusted by the perception and interaction domain, enabling flexible and controllable movement posture without modifying the underlying program.

[0069] The underlying flexible balance protection mechanism: by default, it autonomously maintains the stability of the machine's center of gravity to safeguard the overall safety of the machine. When the sensing and interaction domain issues special action commands, it can temporarily shield or weaken the balance constraints to meet the needs of complex movements, special postures, and unbalanced working conditions. It is responsible for joint closed-loop control, dynamic adjustment of the center of gravity, and autonomous gait cycle.

[0070] The fault is automatically isolated when a single joint is damaged, the machine stops when the damaged joint is damaged, and the remaining joints continue to operate autonomously to compensate. The fault compensation mechanism is manifested externally as an asymmetrical motion mode (the faulty joint is locked, and the adjacent joints move excessively to maintain the overall function), and the whole machine does not stop running.

[0071] It only receives action commands from the perception and interaction domain and does not communicate directly with the logical cognition domain; this domain is the underlying real-time execution unit, which is only responsible for motion execution and does not participate in data calculation and logical decision-making.

[0072] The logic for three-domain collaborative work is as follows:

[0073] 1. The perception and interaction domain collects environmental images and scene perception information;

[0074] 2. The perception and interaction domain can directly send action commands and motion parameters to the motion control domain, which executes independently based on a deterministic motion algorithm; basic instinctive actions do not require the participation of the logical cognition domain.

[0075] 3. The perception and interaction domain is responsible for environmental perception, rapid decision-making, and issuing action intentions. The explicit instructions are actions such as walking, stopping, squatting, lifting legs, and single-limb control, and can specify balanced or unbalanced execution modes. After receiving the instructions, the motion control domain autonomously completes the execution of motion details, gait, and posture according to the corresponding mode.

[0076] 4. When faced with complex scenarios and working conditions requiring deep logical operations, the perception and interaction domain calls the logical cognition domain to carry out high-order deep reasoning and algorithm calculations;

[0077] 5. After the logical cognition domain completes the calculation, it feeds back the optimal strategy solution. The perception and interaction domain makes the final decision confirmation and then issues instructions to the motion control domain to complete the execution of the whole machine's limb movements and postures.

[0078] 6. A dual-path collaborative mechanism of autonomous shallow thinking and on-demand deep reasoning:

[0079] Path 1 (Shallow Thinking, Autonomous Decision-Making): The perceptual interaction domain evaluates the current scenario through its internal quick-thinking knowledge base. For daily, repetitive, known, or structured scenarios, the knowledge base can independently complete shallow reasoning and direct decision-making without waking up or calling the logical cognitive domain, resulting in a response time far shorter than the deep reasoning path.

[0080] Path Two (Deep Reasoning, Verification and Storage): Only when the QuickThink Knowledge Base determines that the current scenario exceeds its shallow capabilities (such as unfamiliar environments, complex logic, unknown objects) will the perceptual interaction domain call the logical cognition domain for deep reasoning. Once the reasoning result is verified to be correct and valid, the conclusion is distilled, compressed, and permanently stored in the QuickThink Knowledge Base, transforming the results of deep thinking into shallow experience that can be directly used later.

[0081] Frequency of logical cognition domain invocation: As the running time increases and the more scenarios the QuickThink knowledge base covers, the frequency of logical cognition domain invocation decreases significantly, ultimately achieving the ideal running state where "the vast majority of daily tasks are autonomously decided by the perceptual interaction domain at a shallow level, and a very small number of complex tasks are deeply participated in by the logical cognition domain".

[0082] This application also features a three-level progressive reasoning chain.

[0083] This system supports the following three-level inference chain, progressing step by step to ensure that there is a processing path in any scenario:

[0084] Level 1 (QuickThink Knowledge Base): For everyday known scenarios, the QuickThink Knowledge Base directly matches experience and makes decisions for execution, providing instantaneous response while the logical cognitive domain remains in low-power sleep mode.

[0085] Level 2 (Local Logical Cognitive Domain): When the QuickThink knowledge base determines that the scenario is unfamiliar or lacks sufficient experience, the perception interaction domain calls the local logical cognitive domain to perform deep reasoning.

[0086] Level 3 (Cloud-based Master Machine Collaboration): When the local logical cognitive domain fails to reason due to insufficient computing power, incomplete model coverage, or other reasons, the perceptual interaction domain attempts to upload the task to the cloud master machine for collaborative reasoning. After the cloud-based reasoning is completed, the result is sent back to the perceptual interaction domain.

[0087] The triggering of cloud-based master-machine collaboration requires the simultaneous fulfillment of two conditions: ① the local logical cognitive domain cannot complete reasoning; ② the network link is available. If the network is unavailable, the system automatically switches to a conservative operating mode (pausing complex tasks while maintaining basic perception and motor capabilities), without affecting the integrity and independence of core functions.

[0088] Regardless of whether the reasoning result comes from the second or third level, after being verified as correct by the perceptual interaction domain, it undergoes knowledge distillation and is stored in the QuickThink knowledge base, transforming it into local experience for direct subsequent use. The cloud-based master machine can simultaneously serve multiple robot terminals, and its experience base can be synchronously shared with terminals of the same model, realizing the collective evolution capability of "one machine learns, the whole network reuses".

[0089] It should be noted that cloud-based master-machine collaboration is an optional enhancement and not a necessary component of this system. Systems without cloud connectivity will automatically switch to a conservative operating mode (pausing complex tasks while maintaining basic capabilities) when second-level inference fails, without affecting the integrity and independence of the system's core functions.

[0090] This application features a dynamic hierarchical power supply and inference motion power consumption interlocking mechanism: the total power supply resources of the whole machine are fixed, and the power supply and computing power resources of each domain are actively and dynamically allocated by the perception and interaction domain according to the current operating needs of the whole machine.

[0091] The system also includes a daily standby state, a deep thinking-complex logic reasoning state, and a high-speed movement-complex movement state;

[0092] In normal standby mode, the perception and interaction domain unit controls the logic and cognition domain unit to reduce the main frequency of operation, reduce task overhead, or cut off the power supply of other domains, so that the machine enters a low-power sleep operation state, or directly shuts down the power supply of the functional module, so that the whole machine enters an energy-saving state.

[0093] Deep Thinking - The state of complex logical reasoning is controlled by the perception and interaction domain, which improves the power supply priority of the logical cognition domain, so that the reasoning computing power of the logical cognition domain reaches the full load state. At the same time, the motion control domain is constrained to retain only the basic standing balance function, so as to realize the staggered scheduling of reasoning computing power power consumption and motion control power consumption, and avoid the failure of the whole machine power consumption exceeding the standard and overload caused by the simultaneous full load of the two domains.

[0094] High-speed motion and complex motion states are controlled by the perception and interaction domain, which reduces the power supply quota of the logic and cognition domain, suspends deep reasoning tasks, and prioritizes the energy supply of the motion control domain to ensure stable motion and a stable center of gravity that prevents the machine from tipping over.

[0095] During continuous operation, external power supply monitoring can observe a characteristic pattern of alternating high-power pulses (logic inference) and stable power consumption (motion or standby).

[0096] The system employs a layered, independent heat dissipation structure. The perception and interaction domain utilizes a high-load active heat dissipation structure. The logic and cognition domain exhibits intermittent, pulsed heat generation and utilizes an independent heat dissipation structure adapted to intermittent loads. The motion control domain prioritizes overall low power consumption and low heat generation, employing a simple heat dissipation structure adapted to low loads. Each of the three domains operates with independent heat dissipation, preventing heat crosstalk and ensuring tiered heat dissipation based on load characteristics. This guarantees stable, long-term operation under high loads without frequency throttling or stuttering.

[0097] The three domains have distinct appearances in terms of heat dissipation structure: the perception and interaction domain has dense heat dissipation holes and is equipped with an active fan; the logic and cognition domain has sparse heat dissipation holes (suitable for pulse cooling); and the motion control domain has a simple heat sink or no heat dissipation openings.

[0098] This system has the advantage of modular upgrades:

[0099] 1. This system can improve logical reasoning ability: by upgrading the logical cognitive domain reasoning algorithm and computing hardware separately;

[0100] 2. This system can enhance environmental awareness and interaction capabilities: it only expands the local storage of the perception and interaction domain and the scene knowledge base;

[0101] 3. This system can improve limb movement ability: only the gait and posture control algorithms in the motion control domain are optimized.

[0102] Each domain can be independently upgraded and replaced or optimized without modifying the other two domain control systems, making it highly adaptable and maintainable.

[0103] Optional configuration details:

[0104] In this architecture, the logical cognition domain is an optional configuration module. In a reduced-configuration state (i.e., without deploying the logical cognition domain), the perception and interaction domain and the motion control domain can still constitute a complete, streamlined two-domain system. In this case, the perception and interaction domain independently completes autonomous decision-making and execution of daily tasks through the built-in QuickThink knowledge base. The robot still possesses environmental perception, shallow thinking, and real-time motion capabilities, making it suitable for scenarios with clearly defined task boundaries, cost sensitivity, or where deep reasoning is not required. Developers (or users) can subsequently add the logical cognition domain module at any time to upgrade the system to a complete three-domain system, and the existing experience data in the QuickThink knowledge base can be seamlessly integrated, achieving "plug and play, leap in capabilities."

[0105] In the three-domain hierarchical architecture of this invention, the perception and interaction domain (including the QuickThink knowledge base) and the motion control domain constitute the core basic units of the robot. Both are indispensable and together endow the robot with the basic intelligent characteristics of a "living thing"—that is, the ability to perceive the environment, make autonomous decisions based on shallow thinking, and execute movements in real time.

[0106] Based on this, modular tailoring can be carried out according to product positioning and cost requirements:

[0107] This includes the full configuration (three-domain full-featured version):

[0108] Simultaneously deploying the perception and interaction domain, logical cognition domain, and motion control domain, and adding deep reasoning capabilities on top of the core basic unit, it realizes a complete closed-loop capability of "shallow thinking + on-demand deep reasoning + real-time motion", which is suitable for high-end products such as humanoid robots and fully autonomous mobile robots.

[0109] This system can be implemented as a simplified dual-domain version (without deep reasoning capabilities):

[0110] Only the perception and interaction domain (including the QuickThink knowledge base) and the motion control domain are retained, while the logical cognition domain is reduced. Under this configuration: the robot still possesses complete environmental perception, shallow thinking autonomous decision-making, and real-time motion execution capabilities; it can independently handle all daily, known, and structured scenario tasks (such as obstacle avoidance, following, and standard interaction); it lacks complex logical deep reasoning capabilities and cannot handle unfamiliar, abstract, or highly complex tasks; it is suitable for scenarios with clear task boundaries, cost sensitivity, or no need for deep reasoning (such as fixed route inspection, basic welcoming robots, and educational kits); the QuickThink knowledge base in the perception and interaction domain can still continuously accumulate experience (through user manual annotation or cloud synchronization), but the generation of new experience requires external tools or human intervention.

[0111] This system has a progressive upgrade path:

[0112] The same robot body can support plug-and-play hardware upgrades from a dual-domain simplified version to a three-domain complete version: after users or developers add the logic cognition domain module, the whole machine will automatically gain deep reasoning ability, and the experience data in the original QuickThink knowledge base can be seamlessly connected to the new system to achieve "plug and play, and leap in ability".

[0113] In summary: the adoption of a biomimetic three-domain hierarchical architecture design makes the robot's intelligent behavior logic more in line with the characteristics of biomimetic hierarchical collaboration;

[0114] By using a three-domain computing power layered decoupling and dynamic power consumption interlocking mechanism, idle computing power is managed in a low-power sleep mode, which significantly reduces the overall power consumption of the device and improves the terminal's battery life.

[0115] The system adopts a three-domain independent heat dissipation and power consumption staggered operation design, which significantly reduces the heat load of the whole machine and avoids overheating, frequency reduction and crashes caused by high heat generation during logic reasoning and physical movement at the same time.

[0116] The hardware and algorithms in each domain are modularly and independently laid out, which facilitates technology iteration, maintenance and repair, and large-scale mass production.

[0117] It has the ability to autonomously isolate and compensate for joint failures, enabling local fault-tolerant operation and ensuring continuous and uninterrupted operation of the entire machine.

[0118] Through dynamic and intelligent allocation of resources across three domains, it can flexibly adapt to robot models of different configuration levels, product positioning, and any form, including but not limited to legged, wheeled, tracked, humanoid, bionic animal, industrial robotic arms, composite robots, and other bionic or non-bionic configurations.

[0119] This invention breaks the technical bias of "computing power redundancy preference" in the field and achieves Pareto improvement in system-level power consumption by actively reducing the power supply and computing power of unnecessary domains.

[0120] Significantly improved real-time interaction capabilities: The QuickThink knowledge base supports extremely low-latency responses to known scenarios, greatly reducing the robot's execution latency for daily and repetitive tasks, and the smoothness of human-computer interaction is significantly better than that of a pure inference-driven architecture; at the same time, it further reduces the frequency of wake-up in the logical cognition domain, and the average power consumption of the whole machine is further optimized on the basis of the existing interlocking mechanism.

[0121] This invention is an architecture-level invention. Features such as three-domain layering, QuickThink knowledge base, and power consumption interlocking are all definitions at the system architecture level. It does not limit the specific communication protocol, similarity matching algorithm, experience data storage format, or specific implementation method of power supply scheduling.

[0122] Example 2:

[0123] like Figures 1 to 3As shown, a biomimetic three-domain hierarchical decoupling and dynamic power consumption interlocking robot intelligent control system is presented.

[0124] The hardware implementation of this system is as follows:

[0125] 1. Perception and Interaction Domain: Equipped with a high-performance main control processing unit, external binocular vision acquisition module, LiDAR and various body sensing units, and equipped with a large-capacity local storage unit;

[0126] 2. Logical cognitive domain: Equipped with a dedicated artificial intelligence computing power reasoning unit, it has the ability to perform offline logical deduction and deep learning operations on large parameter models;

[0127] 3. Motion control domain: Equipped with a real-time motion control processing unit and a multi-joint drive control module, it realizes closed-loop control of body posture and motion;

[0128] 4. Power supply section: It adopts an intelligent power management system that supports dynamic voltage and current regulation. The sensing and interaction domain actively manages the power supply strategy and power consumption scheduling based on real-time operating needs.

[0129] The software implementation of this system is as follows:

[0130] Perception and Interaction Domain Software: Equipped with a multi-tasking operating system, it runs multi-sensor perception fusion, scene intelligent decision-making, and whole-machine power management and scheduling programs; it also has a built-in QuickThink knowledge base management software module, which supports distilled storage of experience data, vectorized index construction, fast similarity matching query, and a result synchronization update mechanism with the logical cognition domain.

[0131] Logical cognitive domain software: Configured with a dedicated artificial intelligence reasoning framework to support high-order logical operations, task inference, and intelligent decision-making algorithms;

[0132] Motion control domain software: Equipped with a real-time embedded operating system, it runs gait planning, posture balance control, and joint closed-loop drive control algorithms.

[0133] The specific implementation methods described above, including but not limited to vectorized indexing and similarity matching, are used to illustrate the feasibility of this invention. This invention is an architecture-level invention and is not limited to the specific implementation methods described above. Any system that possesses the functional division and collaboration mechanism described in this architecture falls within the scope of protection of this patent.

[0134] The workflow implementation example of this application is as follows:

[0135] 1. The robot is powered on and started up. The perception and interaction domain completes the initialization of all modules, and the logical cognition domain enters a low-power standby mode.

[0136] 2. The perception and interaction domain collects environmental information through various sensors, and first conducts autonomous, shallow thinking and decision-making using the internal QuickThink knowledge base:

[0137] If the knowledge base determines that the current scene can be covered by existing experience, it directly generates action instructions and motion parameters and sends them to the motion control domain for execution, while the logic cognition domain remains in low-power sleep mode throughout the process.

[0138] If the knowledge base determines that the scenario is unfamiliar, complex, or beyond the scope of shallow reasoning, it proceeds to step 3 to invoke the logical cognitive domain.

[0139] 3. When encountering complex tasks, such as mathematical operations and high-order logical reasoning scenarios, the perception and interaction domain schedules the logical cognition domain to operate at full power, while constraining the motion control domain to move significantly, retaining only basic balance capabilities.

[0140] When the logical cognition domain reasoning fails in step 3, the perception interaction domain automatically determines that the local reasoning ability is insufficient and starts the cloud master machine collaborative mode: uploading the current task data to the cloud master machine server.

[0141] After the cloud-based master machine completes the reasoning using its enhanced computing power, it sends the strategy results back. Upon receiving the results from the cloud, the perception and interaction domain executes and verifies them. Once verified, the results are distilled and stored in the QuickThink knowledge base, and the experience is marked as "verified" in the cloud-based master machine's experience base.

[0142] The cloud-based master robot experience database can be periodically synchronized to each robot terminal to achieve group experience sharing.

[0143] 4. After completing the deep reasoning operation, the logical cognition domain sends the optimal strategy result back to the perceptual interaction domain. The perceptual interaction domain executes the strategy and verifies its effectiveness.

[0144] If the verification passes (task successfully completed), the perceptual interaction domain will perform knowledge distillation and compression on the conclusion of the strategy and store it in the QuickThink knowledge base, transforming the results of this deep reasoning into shallow experience that can be directly called in the future.

[0145] If verification fails, the data will not be stored in the knowledge base, and a decision will be made based on the failure information to either re-invoke the logical cognitive domain or adopt a degradation strategy.

[0146] Subsequently, the sensory interaction domain sends instructions to the motor control domain to execute the corresponding limb movements.

[0147] 5. After the complex reasoning task is completed, the perception and interaction domain control logic cognitive domain falls back to low-power standby mode, while the motion control domain restores its normal motion capability under all working conditions.

[0148] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A biomimetic three-domain hierarchical decoupling and dynamic power consumption interlocking robot intelligent control system, characterized in that, include: The system consists of three physically separate computing domains: a logical cognition domain unit, a perception and interaction domain unit, and a motion control domain unit. Each computing domain dynamically allocates electrical energy. The logical cognition domain unit is responsible for global cognitive reasoning and task planning, the perception and interaction domain unit is responsible for environmental perception and data preprocessing, and the motion control domain unit is responsible for distributed motion execution and local closed-loop control. The perception and interaction domain unit is the core of the overall control of the machine, coordinating the collaborative work of the logical cognition domain unit and the motion control domain unit.

2. The biomimetic three-domain hierarchical decoupling and dynamic power consumption interlocking robot intelligent control system as described in claim 1, characterized in that, The perception interaction domain unit includes a main control module, a perception module, and a storage module; The main control module of the whole machine is the only main control unit of the whole machine. It coordinates the communication, task priority and mode switching of all modules. The main control module of the whole machine dynamically configures the motion parameters of the motion control domain and can directly issue single limb control commands. It can also issue commands to the motion control domain to temporarily shield the flexible balance protection mechanism to execute special actions. The perception center module is responsible for environmental perception and scene recognition, and the main control center module of the whole machine makes scene decisions based on the perception results. The storage module is an expandable, high-capacity local storage module for storing scene knowledge base, visual features, and environmental memory; The perception and interaction domain unit undertakes the functions of power distribution, power management, dynamic power consumption scheduling and interlock management of the whole machine.

3. The biomimetic three-domain hierarchical decoupling and dynamic power consumption interlocking robot intelligent control system as described in claim 1, characterized in that, The logical cognitive domain unit includes a logical algorithm subunit and a deep learning reasoning dedicated unit. The logical cognitive domain unit is used for mathematical algorithms, logical deduction, deep learning of complex tasks, abstract theoretical calculation, and high-order strategy reasoning. There is no direct data link between the logical cognitive domain unit and the motion control domain unit. All data interactions are relayed through the perception interaction domain unit, and the logical cognitive domain unit refuses to execute any instructions from the motion control domain unit.

4. The biomimetic three-domain hierarchical decoupling and dynamic power consumption interlocking robot intelligent control system as described in claim 1, characterized in that, The motion control domain unit adopts a distributed architecture, and each joint drive subunit has independent drive control and state feedback capabilities, which can realize local closed-loop control of a single joint.

5. The biomimetic three-domain hierarchical decoupling and dynamic power consumption interlocking robot intelligent control system as described in claim 1, characterized in that, The perceptual interaction domain unit is also equipped with a QuickThink knowledge base subsystem to enable autonomous shallow thinking and rapid decision-making capabilities, thus eliminating the reliance on deep reasoning in the logical cognition domain. Once the perception and interaction domain acquires environmental features, the QuickThink knowledge base subsystem is responsible for rapid matching and evaluation. For scenarios where a match is successfully made, the system directly generates control commands and sends them to the motion control domain for execution, achieving an instantaneous response with extremely low latency. When the QuickThink knowledge base subsystem determines that the current scene exceeds its shallow reasoning ability, the perception interaction domain unit calls the logical cognition domain unit to perform deep reasoning. The QuickThink knowledge base subsystem has a verification and storage closed loop: after the logical cognitive domain reasoning result is verified to be correct, it is distilled and stored in the QuickThink knowledge base; The QuickThink knowledge base subsystem can evolve sustainably: it thinks when it encounters something new, learns when it thinks, and applies what it learns immediately.

6. The biomimetic three-domain hierarchical decoupling and dynamic power consumption interlocking robot intelligent control system as described in claim 5, characterized in that, The perception and interaction domain unit actively and dynamically allocates power supply and computing resources to each domain based on the current overall machine operation requirements and the fixed total energy resources of the overall machine.

7. The biomimetic three-domain hierarchical decoupling and dynamic power consumption interlocking robot intelligent control system as described in claim 6, characterized in that, The system includes a daily standby state, a deep thinking-complex logic reasoning state, and a high-speed movement-complex movement state; In the normal standby state, the perception and interaction domain unit controls the logic cognition domain unit to reduce the main frequency of operation and reduce task overhead, so that the machine enters a low-power sleep operation state, or directly shuts off the power supply of the functional module, so that the machine as a whole enters an energy-saving state. The deep thinking-complex logic reasoning state is controlled by the perception interaction domain, which increases the power supply priority of the logic cognition domain, so that the reasoning computing power of the logic cognition domain reaches the full load state. At the same time, the motion control domain is constrained to retain only the basic standing balance function, so as to realize the staggered scheduling of reasoning computing power power consumption and motion control power consumption, and avoid the failure of the whole machine power consumption exceeding the standard and overload caused by the simultaneous full load of the two domains. The high-speed motion-complex motion state is controlled by the perception and interaction domain, which reduces the power supply quota of the logic and cognition domain, suspends deep reasoning tasks, and prioritizes the supply of the whole machine's energy to the motion control domain to ensure stable body movement and a stable center of gravity and posture so as not to tip over.

8. The biomimetic three-domain hierarchical decoupling and dynamic power consumption interlocking robot intelligent control system as described in claim 1, characterized in that, The system adopts a layered independent heat dissipation structure; The sensing interaction domain adopts a high-load active heat dissipation structure; The logical cognitive domain exhibits pulsed intermittent heat generation and employs an independent heat dissipation structure adapted to intermittent loads. The motion control domain features low overall power consumption and low heat generation, employing a simple heat dissipation structure adapted to low loads. The three-domain heat dissipation structures are independent of each other, and heat is not cross-exposed to each other, with heat dissipation in stages according to load characteristics.

Images