Methods, apparatus, equipment, and media for secure compilation and verification of robot swarms
By employing a secure compilation and verification method for heterogeneous robot clusters, and using a three-layer verification approach involving differentiated preprocessing at the control layer and edge execution modules, a deployment package carrying a signature is generated, forming a dual closed-loop protection. This addresses the shortcomings in security and flexibility of traditional security protection models, and achieves highly reliable security strategy iteration and equipment operation reliability.
Patent Information
- Application Number
- CN202611332331.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-31
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2046-08-31
AI Technical Summary
The current traditional safety management and protection model for heterogeneous robot clusters has insufficient overall security, poor protection integrity and strategy iteration flexibility, and cannot adapt to the safety management and control requirements of highly reliable and highly dynamic intelligent cluster operations. In addition, the traditional safety protection system has a single protection dimension and lacks a full-process, closed-loop safety management and control logic, which poses safety risks of equipment malfunction and operational failure.
By employing a secure compilation and verification method, a deployment package carrying a compiler signature or verification digest is generated through a three-layer progressive mandatory verification process involving differentiated preprocessing at the control layer and edge execution modules. Combined with secure configuration data, this enables pre-verification and edge-side verification of control intent data, forming a dual closed-loop protection mechanism to ensure rapid iteration of security policies and reliable operation of devices.
It improves the safety and reliability of the robot cluster and the closed-loop protection, enables rapid iteration of safety strategies and scenario adaptation, eliminates the risk of equipment loss of control, and improves the flexibility and reliability of the overall protection system.
Smart Images

Figure CN122861786B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of robot control technology, and in particular to a secure compilation and verification method, apparatus, device, and medium for robot swarms. Background Technology
[0002] With the continuous iteration and popularization of robot swarm collaborative control and artificial intelligence automated scheduling technology, the heterogeneous robot swarm management and control system has gradually achieved intelligent and digital upgrades. It can adapt to complex scenarios of multi-type robot collaborative operations and has the industry-wide characteristics of flexible scheduling, strong adaptability and high degree of automation. As a result, it has formed the traditional safety management and protection method for the edge execution of heterogeneous robot swarms at this stage, which is widely used in various robot swarm operation safety protection scenarios.
[0003] However, current traditional safety management and protection models for heterogeneous robot swarms still have many technical shortcomings. Overall security, protection integrity, and policy iteration flexibility are all significantly inadequate, failing to meet the high-reliability, high-dynamic intelligent swarm operation safety management requirements. In application scenarios involving intelligent policy generation and equipment scheduling, robot control commands generated by various automated and intelligent methods are uncertain, easily producing abnormal command data that does not meet equipment capability thresholds and physical safety constraints. This results in poor fault tolerance and safety risks such as equipment malfunction and operational failures. Furthermore, existing safety protection systems have a single protection dimension, relying heavily on passive monitoring and interception intervention during the operation phase. They lack a full-process, closed-loop safety management logic, resulting in significant vulnerabilities and difficulty in guaranteeing safety reliability. In addition, the policy update mechanism of traditional safety management models is rigid and inflexible. Iterative optimization of safety protection rules heavily depends on system code modifications and overall deployment updates, leading to long policy adjustment cycles and low efficiency. This makes it impossible to quickly adapt to complex and ever-changing operational scenarios, severely limiting the intelligent safety upgrade and scenario adaptation capabilities of heterogeneous robot swarm management systems. Summary of the Invention
[0004] Therefore, it is necessary to provide a secure compilation and verification method, apparatus, device, and medium for robot swarms to address the aforementioned technical problems.
[0005] A secure compilation and verification method for robot swarms, the method being applied to a heterogeneous robot swarm security management system including a control layer and an edge execution module, the method comprising: Acquire control intent data generated by the control layer, and perform differentiated preprocessing based on the source type of the control intent data, and output the original control intent data or a limited intermediate representation containing only capability-level abstract descriptions. The control intent data or restricted intermediate representation is subject to a three-layer progressive mandatory verification on the control side. After the verification is passed, a deployment package carrying a compiler signature or verification digest is compiled and generated. After atomic verification by the internal components of the deployment package, the deployment package is sent to the edge execution module via the communication bus. The internal components of the deployment package include task logic, version number, and security configuration data that matches the local capability list of the edge execution module. The edge execution module performs edge-side verification on the received deployment package. If the verification passes, all components in the deployment package are loaded and activated to carry out safety management and control operations for the heterogeneous robot cluster. Otherwise, the deployment package is refused to be loaded, and an anomaly is reported to the management layer to trigger the regeneration of control intent data.
[0006] In one embodiment, differential preprocessing is performed based on the source type of the control intent data, resulting in either the original control intent data or a limited intermediate representation containing only a capability-level abstract description, including: For control intent data derived from manual programming or automated tools, remain unchanged; For control intent data automatically generated by the nondeterministic strategy generator, the adapter parses the intent tags and maps each intent tag to a registered skill identifier in the capability whitelist in the edge execution module. The qualifiers in the control intent data are also mapped to preset parameter levels. Based on the mapping results, a restricted intermediate representation is generated. The restricted intermediate representation includes the skill identifier, parameter level, target reference, and preconditions of the capability level, but does not contain underlying motor control commands or direct hardware operation instructions.
[0007] In one embodiment, the three-layer progressive forced verification includes: The first layer of structure verification is used to verify whether the node or skill identifier referenced by the control intent data or restricted intermediate representation exists and is activated in the capability whitelist of the edge execution module; The second layer of parameter constraint verification is used to verify the parameters corresponding to the nodes or skill identifiers referenced by the control intent data or restricted intermediate representation item by item. This includes verifying whether the data type of the parameter is consistent with the data type defined in the capability whitelist, whether the parameter value is within the legal range defined in the capability whitelist, and whether the parameter enumeration value is a legal enumeration value defined in the capability whitelist. The third-level sequence invariant verification is used to verify whether the action sequence corresponding to the control intent data or restricted intermediate representation violates physical safety constraints based on the device security event log and the preset sequence-level physical safety invariant. If any layer of verification fails, the deployment package compilation is rejected, a verification failure event or structured feedback message is generated and reported to the management layer, which then regenerates the control intent data. If the three-layer progressive mandatory verification passes, a deployment package carrying a compiler signature or verification digest is compiled and generated.
[0008] In one embodiment, the sequence-level physical safety invariants include: a speed limit constraint defining that the speed parameter of a single action does not exceed the physical limit of the device; a geofence constraint defining that the target coordinates are within the geofence polygon range that the device is allowed to operate in; a power budget constraint defining that the estimated total power consumption of the action sequence does not exceed the current power of the device minus the minimum power required for a safe return; an attitude constraint defining that the device attitude change required by the action sequence does not exceed the safe range under the current terrain conditions; and a dangerous action cooling cycle constraint defining that a certain high-risk action must not reappear in the current action sequence within the minimum cooling interval after the last execution.
[0009] In one embodiment, the atomic verification of the components within the deployment package includes: verifying whether the task logic, version number, and security configuration data matching the local capability list of the edge execution module in the deployment package all exist and are compatible with each other; if any component fails verification, the deployment package is rejected and an event is generated and reported to the management layer, which then regenerates the control intent data; if all components are verified successfully, the deployment package is sent to the edge execution module via the communication bus.
[0010] In one embodiment, the security configuration data includes a set of security contracts and a data valve configuration; The set of security contracts includes a security contract bound to each control interface that is allowed to be remotely invoked before the control interface of the robot device is allowed to be remotely invoked. Each security contract contains a TTL field and a security degradation action identifier field bound to the corresponding control interface, as well as an edge veto right trigger condition field bound to a single control interface or the device's global security domain. The security contract is used by the security daemon thread in the edge execution module, which is independent of the task execution thread, to perform TTL timeout monitoring of the corresponding control interface at a higher frequency than the task execution thread based on the TTL field, and to perform veto right monitoring of the corresponding control interface at the highest priority based on the edge veto right trigger condition field. When a TTL timeout is detected and the edge hardware-level veto right is triggered, the security degradation action of the corresponding control interface is autonomously controlled according to the security degradation action identifier field without relying on network communication with the management layer. At the same time, the task execution thread is notified to interrupt or cancel the current task logic, an event record is generated and written to the local security event log and / or reported to the management layer. The data valve configuration is used by the data valve execution unit in the edge execution module after being issued to and activated. The data valve performs semantic transformation on the raw sensor data uploaded by the device through a bypass data channel independent of the communication bus, generates semantic key-value pairs, and applies four-dimensional hard constraints in the data valve configuration to filter them. Semantic key-value pairs that satisfy all hard constraints are obtained, and the task execution thread consumes the latest semantic key-value pairs that are consistent with the version number of the deployment package to execute the task logic. The four-dimensional hard constraints include the preset maximum field number limit constraint, the preset maximum byte number limit constraint, the preset maximum output frequency limit constraint, and the preset maximum conversion delay limit constraint for semantic key-value pairs.
[0011] In one embodiment, edge-side verification includes: The deployment package verification unit in the edge execution module verifies whether the compiler signature or verification digest in the issued deployment package exists, is damaged, or has been tampered with, or whether the signature algorithm is mismatched, and verifies whether the version number of the deployment package is compatible with the local capability list.
[0012] A secure compilation and verification device for robot swarms, the device being applied to a heterogeneous robot swarm security management system including a control layer and an edge execution module, the device comprising: The differential preprocessing unit is used to acquire control intent data generated by the control layer and perform differential preprocessing according to the source type of the control intent data, and output the original control intent data or a limited intermediate representation containing only capability level abstract descriptions. The control-side verification unit is used to perform a three-layer progressive mandatory verification on the control intent data or restricted intermediate representation. After the verification is passed, a deployment package carrying a compiler signature or verification digest is compiled and generated. After the atomic verification of the internal components of the deployment package, the deployment package is sent to the edge execution module via the communication bus. The internal components of the deployment package include task logic, version number, and security configuration data that matches the local capability list of the edge execution module. The edge-side verification unit is used by the edge execution module to perform edge-side verification on the received deployment package. If the verification is successful, all components in the deployment package are loaded and activated to carry out safety management and control operations for the heterogeneous robot cluster; otherwise, the deployment package is refused to be loaded and an exception is reported to the management and control layer to trigger the regeneration of control intent data.
[0013] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the above-described secure compilation and verification method for robot swarms.
[0014] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described secure compilation and verification method for robot swarms.
[0015] The aforementioned secure compilation and verification methods, devices, equipment, and media for robot swarms rely on a two-tiered security architecture: pre-compilation verification on the control side and edge-side verification. This architecture performs differentiated preprocessing on multi-source control intent data, uniformly transforming non-standard control intents into standardized, restricted intermediate representations. This avoids the risk of device loss of control caused by the direct issuance of non-standard control intents from the source. The control side employs a three-layer progressive mandatory verification system, distinct from traditional runtime passive interception protection, to achieve unavoidable security review before control intents are issued, preventing the risk of abnormal direct control transmission due to protection mechanism failures. The verification results include a compiler signature or verification digest. The deployment package is fixed in form, and atomic verification of the components within the package completes the pre-hardening, building a pre-security foundation of static mandatory verification + signature solidification. The edge side verifies the entire deployment package before it can be loaded and activated for operation. This forms a dual-locked protection loop of compilation interception on the control side and signature interception on the edge side, which makes up for the defects of traditional security guardrail link breakage. At the same time, the deployment package embeds independent security configuration data, and security rule iteration can be achieved by updating the deployment package without the need to completely reconstruct and reinstall the verification program. This frees the system from the constraints of the release cycle on security policy updates, and takes into account the security of robot cluster operation, the reliability of the protection loop, and the flexibility of security policy iteration. Attached Figure Description
[0016] Figure 1 This is a flowchart illustrating a secure compilation and verification method for a robot cluster in one embodiment; Figure 2 This is a schematic diagram of the deployment package structure belonging to the same version unit in one embodiment; Figure 3 This is a structural block diagram of a secure compilation and verification device for a robot cluster in one embodiment; Figure 4 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation
[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0018] In one embodiment, such as Figure 1 As shown, a secure compilation and verification method for robot swarms is provided. This method is applied to a heterogeneous robot swarm security management system that includes a control layer and an edge execution module, and includes the following steps: Step 1: Obtain the control intent data generated by the control layer, and perform differentiated preprocessing according to the source type of the control intent data, and output the original control intent data or a limited intermediate representation containing only the capability level abstract description.
[0019] It should be understood that the control layer performs differentiated preprocessing on multi-source control intentions based on the source type, converting highly uncertain control intentions into a restricted intermediate representation of capability-level abstract description, thereby blocking the potential for robot equipment to go out of control caused by the direct issuance of non-standard control intentions from the source, while taking into account both data processing efficiency and source safety constraints.
[0020] In this embodiment, differentiated preprocessing is performed based on the source type of the control intent data, resulting in the output of either the original control intent data or a limited intermediate representation containing only a capability-level abstract description, including: The data source type is determined by reading the source identifier field in the header of the control intent data. For control intent data originating from manual orchestration or automated tools, this field remains unchanged. For control intent data automatically generated by non-deterministic policy generators (such as Large Language Model (LLM) / VLA (Vision-Language-Action) models / reinforcement learning policies), the control intent data is input into an adapter. The adapter extracts action descriptions from natural language or semi-structured output to parse intent tags. Each intent tag is mapped to a registered skill identifier in the capability whitelist of the edge execution module (e.g., "move to" is mapped to the `navigate_to` skill in the capability whitelist). Qualifiers in the control intent data are mapped to preset parameter levels (e.g., "fast" is mapped to high speed, "cautious" to low speed; if no qualifier is specified, the default level is used). Based on the mapping results, a restricted intermediate representation is generated. This restricted intermediate representation includes the skill identifier at the capability level, parameter level, target reference, and preconditions, but does not contain underlying motor control commands or direct hardware operation instructions. In cases where the source identifier field is missing, it is treated as a manually orchestrated source by default.
[0021] Specifically, the restricted intermediate mode has several hard constraints. First, the parameter set cannot contain PWM (Pulse Width Modulation) values, speed values, current values, torque values, motor numbers, or any other fields that can directly control the hardware, and cannot carry underlying motor control-related instructions. Second, the task skill identifier must be registered, and the content corresponding to the skill field needs to be entered into the target device's capability whitelist. The speed configuration field can only use three types of enumerated identifiers: low speed, medium speed, and high speed. Specific speed values such as meters per second or kilometers per hour cannot be entered. The actual speed value is determined by the edge execution module within the capability boundaries, based on the safety contract. Constraint parsing generation; the content filled in the target field must be parsed based on the location and resource mapping table in the equipment capability list and converted into real coordinates or resource numbers; each judgment condition in the precondition set must rely on the status data of the equipment's local sensors, power metering devices, etc., and complete the authenticity verification during the program execution phase; the restricted intermediate representation syntax system does not support unconditional jump instructions, there are no jump-related instructions, and all task switching operations need to rely on preconditions to perform judgment processing at runtime; the value range of all numerical parameters in the parameter set must fall within the legal range predefined in the capability whitelist for the corresponding skill.
[0022] It is important to note that once the output of the nondeterministic policy generator is converted into a restricted intermediate representation that satisfies the above constraints by the adapter, the process can proceed to the next step. This conversion process is performed solely based on the format constraints of the restricted intermediate representation and does not depend on an understanding of the internal workings of the nondeterministic policy generator. If the conversion fails (e.g., the intent label cannot be mapped to any registered skill), compilation is rejected directly, an adapter conversion failure event containing a summary of the failure reason and the original input is generated, and this event is fed back to the nondeterministic policy generator to regenerate the control intent data.
[0023] Step 2 involves performing a three-layer progressive mandatory verification on the control intent data or restricted intermediate representation. After the verification passes, a deployment package carrying a compiler signature or verification digest is compiled and generated. After atomic verification by the internal components of the deployment package, the deployment package is sent to the edge execution module via the communication bus. The internal components of the deployment package include task logic, version number, and security configuration data that matches the local capability list of the edge execution module.
[0024] It should be understood that the three-layer progressive mandatory verification executed by the control side is a static rigid review before distribution. No control intent data or restricted intermediate representation can bypass this mandatory verification to directly reach the edge execution module. Unlike the traditional runtime dynamic interception mode, there is no risk of protection bypass failure. Combined with the compiler signature or verification digest of the deployment package and the atomic verification of internal components, the encapsulation and solidification are completed, and the security verification result is bound to the deployment package in an immutable manner, so as to realize the solidification and traceability of the verification logic.
[0025] In this embodiment, the three-layer progressive mandatory verification includes: The first-level structure verification is used to verify whether the nodes or skill identifiers referenced by the control intent data or restricted intermediate representation exist and are active in the capability whitelist of the edge execution module. If the verification fails, the deployment package compilation is rejected, a structure verification failure event (containing a list of nodes / skill identifiers not in the capability whitelist) is generated, and the event is reported to the control layer, which will then correct and generate new control intent data.
[0026] The second layer of parameter constraint verification is used to verify each parameter corresponding to the node or skill identifier referenced by the control intent data or restricted intermediate representation. This includes verifying whether the data type of the parameter is consistent with the data type defined in the capability whitelist, whether the parameter value is within the legal range defined in the capability whitelist, and whether the parameter enumeration value is a legal enumeration value defined in the capability whitelist. If any parameter verification fails, the entire verification fails. At this time, the deployment package compilation is rejected, a parameter constraint verification failure event (including the failed parameter name, actual value, legal range, etc.) is generated and reported to the management layer. After the management layer corrects the error, new control intent data is generated.
[0027] The third-layer sequence invariant verification: Based on device security event logs and preset sequence-level physical security invariants, this verifies whether the action sequence corresponding to control intent data or restricted intermediate representations violates physical security constraints. If the verification fails, the deployment package compilation is rejected, and a structured feedback message is generated (containing the identifier or parameter of the rejected node, the number of the violated security invariant, and a list of suggested alternative skills). Upon receiving the feedback, the control layer can learn "what was rejected" and "what alternative solutions can be tried," thereby adjusting the control intent data to pass this layer's verification. It is important to note that the structured feedback message does not contain the complete set of rules and the internal implementation of the decision algorithm for the third-layer sequence invariant verification at the sequence level. This prevents the control layer or the non-deterministic policy generator from obtaining the complete set of security rules from the feedback, thus preventing reverse engineering of security boundaries through feedback.
[0028] Specifically, the aforementioned sequence-level physical safety invariants include: a speed limit constraint defining that the speed parameter of a single action does not exceed the physical limit of the device; a geofence constraint defining that the target coordinates are within the geofence polygon range that the device is allowed to operate in; a power budget constraint defining that the estimated total power consumption of the action sequence does not exceed the current power of the device minus the minimum power required for safe return; an attitude constraint defining that the device attitude change required by the action sequence does not exceed the safe range under the current terrain conditions; and a dangerous action cooling cycle constraint defining that a specific high-risk action must not reappear in the current action sequence within the minimum cooling interval after the last execution.
[0029] In summary, in the three-layer progressive mandatory verification, if any layer fails, the deployment package compilation is rejected, a verification failure event or structured feedback message is generated and reported to the control layer, which then regenerates the control intent data. If the three-layer progressive mandatory verification passes, a deployment package carrying a compiler signature or verification digest is generated. Specifically, the verification digest is generated by calculating a hash digest of the deployment package content, or by signing it using the compiler's private key.
[0030] After the three-layer progressive mandatory verification passes, further atomic verification is performed on the components within the deployment package, including: verification Figure 2 The task logic, version number, and security configuration data (including the set of security contracts and the configuration of data valves) in the deployment package shown are all present and compatible with each other, matching the local capability list of the edge execution module. Figure 2 The task logic includes the execution node sequence, control flow structure, and target execution domain identifier. The security contract set contains security contracts bound one-to-one with each remotely invoked control interface. Each security contract includes the TTL (Time To Live) field, fallback action identifier field, and veto trigger condition field for the corresponding control interface. The data valve configuration includes the upper limit of semantic output fields, maximum output frequency, maximum byte count, maximum conversion latency, and bypass routing rules. Version numbers and local capability list references include version numbers, capability references, and target domains. The compatibility verification between components includes: the control interfaces referenced in the security contract set actually exist in the task logic and the correspondence is correct; the schema version number in the data valve configuration matches the version number of the deployment package; and the control interfaces in the security contract set have corresponding routing rules in the data valve configuration. If any component verification fails, the deployment package is rejected, and an event is generated and reported to the management layer, which then regenerates the control intent data. During atomic verification, if the version number of the deployment package is incompatible with the version of the local capability list of the edge execution module, the entire deployment package is rejected, generating a version incompatibility event (including the deployment package version number, local version number, and a list of conflicting fields) and reporting it to the control layer. The control layer needs to update the deployment package or device capability list and resubmit it. If the control interface referenced in the security contract set does not exist in the task logic, or the control interface called in the task logic is missing in the security contract set, the entire deployment package is rejected, generating an interface mismatch event (including a list of missing interfaces). In this case, the deployment package needs to be corrected and resubmitted for atomic verification. If all components are verified successfully, the deployment package is sent to the edge execution module via the communication bus.
[0031] In this embodiment, the set of security contracts includes a security contract bound to each control interface that is allowed to be remotely invoked before the control interface of the robot device is allowed to be remotely invoked (i.e., at least one of the device registration stage, device network access stage, capability list loading stage, or deployment package activation stage). Each security contract includes a TTL field and a security degradation action identifier field bound to the corresponding control interface, as well as an edge veto trigger condition field bound to a single control interface or to the device's global security domain. The security contract, after being issued to and activated by the edge execution module, is used by a security daemon thread within the edge execution module, independent of the task execution thread. Based on the TTL field, it performs TTL timeout monitoring of the corresponding control interface at a higher frequency (e.g., ≥200Hz) than the task execution thread. Based on the edge veto trigger condition field, it performs veto monitoring of the corresponding control interface with the highest priority. Upon detecting a TTL timeout and triggering an edge hardware-level veto, it autonomously controls the security degradation action of the corresponding control interface (e.g., sending a zero-speed command / emergency stop / holding the current state / returning to a preset security origin / not performing any action / sending a custom degradation command to a specified topic, etc.) according to the security degradation action identifier field, without relying on network communication with the management layer. Simultaneously, it notifies the task execution thread to interrupt or cancel the current task logic, generates an event record and writes it to the local security event log and / or reports it to the management layer. Specifically, the security daemon thread includes at least one of the following: an independent thread, an independent process, a real-time task, or a security coprocessor task.
[0032] It is important to note that hardware-level veto power has the highest priority and can only be modified through device firmware, trusted underlying security modules, or security update processes verified by the hardware root of trust. It cannot be overridden by any declarative configuration. TTL timeout degradation triggered by the TTL field of the security contract has the next highest priority, and both hardware-level veto power and TTL timeout degradation take precedence over declarative security policies (configurable policies issued with configuration have the lowest priority). This hierarchical priority mechanism forms a layered security defense system. When multiple security risks are triggered simultaneously by the device, hardware-level veto power caused by abnormal hardware sensor conditions is responded to first, followed by TTL timeout degradation caused by expired remote commands. These actions will not be intercepted, delayed, or overridden by low-priority declarative security policies, preventing situations where business configuration policies restrict the execution of underlying emergency security actions. From a scheduling priority perspective, this ensures that device physical security and command timing security always take precedence over business optimization configuration policies, completely avoiding the problem of upper-layer business policies crowding out lower-layer emergency protection resources and masking the risks of dangerous device conditions and expired commands. Furthermore, by constructing a dual-layer parallel security protection system of TTL timeout monitoring and hardware-level veto power: on the one hand, it can rely on the edge local monotonic clock and TTL field to monitor whether the remote control commands received by the control interface have timed out, and independently inspect and intercept timed-out invalid commands; on the other hand, it triggers edge hardware-level veto based on the local sensor operating conditions. At the same time, both risks can be autonomously triggered offline to trigger security degradation, fully covering the two core risks of remote control: command timing failure and abnormal physical operating conditions of equipment. Moreover, security degradation is executed on the edge side. Even if the upper-layer network is completely disconnected, the remote management link fails, or the business task thread is stuck and crashed, monitoring and emergency handling can still continue to run on the edge side. It can also proactively notify the task execution thread to immediately interrupt the existing dangerous task logic and generate a complete structured event record to be stored in the local log or reported to the upper layer, forming a complete security closed loop of pre-event time constraints, in-event hardware backup, and post-event traceability and auditing. This significantly reduces the risk of robot misoperation and equipment damage caused by network fluctuations, disconnections, sensor anomalies, and command delays, and significantly improves the system's offline survivability and functional safety level.
[0033] The data valve configuration is used by the data valve execution unit in the edge execution module after being issued to and activated. This unit performs semantic transformation on the raw sensor data (high-bandwidth data such as video streams, raw point clouds, hyperspectral data; low-bandwidth data such as temperature readings, battery levels) uploaded by the device through a bypass data channel independent of the communication bus. This generates semantic key-value pairs and applies four-dimensional hard constraints in the data valve configuration for filtering. Semantic key-value pairs that satisfy all hard constraints are obtained, and the task execution thread consumes the latest semantic key-value pairs that match the version number of the deployment package to execute the task logic. The four-dimensional hard constraints include the preset maximum number of fields for semantic key-value pairs (e.g., ≤16 key-value pairs), the preset maximum number of bytes (e.g., ≤1KB), the preset maximum output frequency (e.g., ≤10Hz), and the preset maximum conversion delay (e.g., ≤10ms).
[0034] Among them, the preset maximum field number limit constraint is used to limit the number of semantic key-value pairs contained in the total semantic key-value pairs obtained after semantic dimensionality reduction transformation of a single frame of sensor data to not exceed the first preset threshold; the preset maximum byte number limit constraint is used to limit the byte length of the total semantic key-value pairs after serialization to not exceed the second preset threshold; the preset maximum output frequency limit constraint is used to limit the number of semantic key-value pairs injected into the semantic surface namespace by the semantic dimensionality reduction transformation link within a unit time window to not exceed the third preset threshold; and the preset maximum transformation delay limit constraint is used to limit the time consumed by the semantic dimensionality reduction transformation link for semantic dimensionality reduction transformation of a single frame of sensor data to not exceed the fourth preset threshold. When any hard constraint is not satisfied, the hard constraint verification is deemed to have failed and a preset exception handling is executed. At the same time, an exception event is generated and reported to the management layer. When all hard constraints are satisfied, the hard constraint verification is deemed to have passed. The semantic key-value pairs that satisfy all hard constraints, along with their corresponding schema version numbers, are routed to the semantic plane namespace. Finally, when the task execution thread executes the heterogeneous robot cluster control task, it cyclically pulls the latest semantic key-value pairs in the semantic plane namespace. After verifying the consistency between the schema version number bound to the semantic key-value pair and the version number of the deployment package, the semantic key-value pairs that have passed the version verification are consumed to execute the task logic.
[0035] It should be understood that, relying on the data valve configuration, the edge side can uniformly complete the semantic structure transformation of the high-bandwidth sensor raw data uploaded by the bypass channel. By using four types of hard boundary conditions—field quantity, byte size, output frequency, and conversion latency—data filtering and compression are completed, eliminating excessive and redundant sensor data, and retaining only compliant semantic key-value pairs for the task execution thread to call. Furthermore, binding the deployment package version ensures that the data and task versions match, which reduces the bus transmission load, lowers computing power consumption, controls data transmission latency and throughput pressure, and ensures the stable and sequential operation of the robot's task logic.
[0036] Step 3: The edge execution module performs edge-side verification on the received deployment package. If the verification passes, all components in the deployment package are loaded and activated to carry out safety management and control operations for the heterogeneous robot cluster. Otherwise, the deployment package is refused to be loaded, and an anomaly is reported to the management layer to trigger the regeneration of control intent data.
[0037] It should be understood that the edge execution module independently verifies the legitimacy of the deployment package. Only when the verification passes will the task logic, version number, and adapted security configuration data in the deployment package be loaded and run. If the verification fails, loading will be directly rejected and an exception will be sent back to the management layer to reconstruct the control intent. This forms a dual security lock-in loop of compilation blocking on the management side and loading blocking on the edge side, making up for the defects of traditional protection link gaps. Furthermore, relying on the independent security configuration data carried by the deployment package, security rule iteration only requires updating the deployment file, without the need for a complete redesign of the deployment verification program, which greatly shortens the security policy iteration cycle and adapts to the rapid protection upgrade of dynamic cluster scenarios.
[0038] In this embodiment, edge-side verification includes: The deployment package verification unit in the edge execution module verifies whether the compiler signature or verification digest in the issued deployment package exists, is corrupted, or has been tampered with, or whether the signature algorithm is mismatched, and verifies whether the version number of the deployment package is compatible with the local capability list. If the deployment package does not carry a compiler signature, the signature data is corrupted, the content has been tampered with resulting in a signature mismatch, or the signature algorithm is incompatible with the deployment package verification unit, the deployment package is refused to be loaded, a security event (including the reason for refusal, deployment package ID, and timestamp) is generated and reported to the control layer. At this time, the device maintains its current operating state or the operating state defined by the last valid deployment package, waiting for the control layer to decide to recompile and issue the deployment package or for manual intervention. If the capability identifier referenced by the deployment package does not exist in the local capability list of the edge execution module or the version is mismatched, the deployment package is refused to be loaded, a capability missing event (including the missing capability ID, the version required by the deployment package, and the actual local version) is generated and reported to the control layer. At this time, the device maintains its current operating state or the operating state defined by the last valid deployment package, waiting for the control layer to decide to recompile and issue the deployment package or for manual intervention.
[0039] In summary, the secure compilation and verification method for robot clusters provided in this application differentiates various control intent data at the control end, standardizes highly uncertain control intents into capability-restricted intermediate representations, constrains the transmission risks of non-standard hardware description information from the source, and completes pre-security hardening by combining the three-layer progressive mandatory verification, deployment package signature solidification, and component atomic verification on the control side. This differs from the traditional passive protection mode at runtime, avoids protection failure vulnerabilities, and forms a closed-loop protection mechanism with two-end linkage by combining the legality verification of deployment packages at the edge end. This effectively improves the overall operational security of heterogeneous robot clusters. At the same time, the security configuration data is embedded in the deployment package, and the policy iteration does not require a complete reconstruction of the deployment system, which significantly improves the iteration flexibility and scenario adaptability of security policies for robot clusters.
[0040] In one embodiment, such as Figure 3 As shown, a secure compilation and verification device for robot swarms is provided. The device is applied to a heterogeneous robot swarm security management system including a control layer and an edge execution module. The device includes: The differential preprocessing unit is used to acquire control intent data generated by the control layer and perform differential preprocessing according to the source type of the control intent data, and output the original control intent data or a limited intermediate representation containing only capability level abstract descriptions. The control-side verification unit is used to perform a three-layer progressive mandatory verification on the control intent data or restricted intermediate representation. After the verification is passed, a deployment package carrying a compiler signature or verification digest is compiled and generated. After the atomic verification of the internal components of the deployment package, the deployment package is sent to the edge execution module via the communication bus. The internal components of the deployment package include task logic, version number, and security configuration data that matches the local capability list of the edge execution module. The edge-side verification unit is used by the edge execution module to perform edge-side verification on the received deployment package. If the verification is successful, all components in the deployment package are loaded and activated to carry out safety management and control operations for the heterogeneous robot cluster; otherwise, the deployment package is refused to be loaded and an exception is reported to the management and control layer to trigger the regeneration of control intent data.
[0041] Specific limitations regarding the secure compilation and verification apparatus for robot swarms can be found in the limitations of the secure compilation and verification methods for robot swarms described above, and will not be repeated here. Each module in the aforementioned secure compilation and verification apparatus for robot swarms can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.
[0042] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores data. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements a secure compilation and verification method for robot swarms.
[0043] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0044] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to perform the following steps: Acquire control intent data generated by the control layer, and perform differentiated preprocessing based on the source type of the control intent data, and output the original control intent data or a limited intermediate representation containing only capability-level abstract descriptions. The control intent data or restricted intermediate representation is subject to a three-layer progressive mandatory verification on the control side. After the verification is passed, a deployment package carrying a compiler signature or verification digest is compiled and generated. After atomic verification by the internal components of the deployment package, the deployment package is sent to the edge execution module via the communication bus. The internal components of the deployment package include task logic, version number, and security configuration data that matches the local capability list of the edge execution module. The edge execution module performs edge-side verification on the received deployment package. If the verification passes, all components in the deployment package are loaded and activated to carry out safety management and control operations for the heterogeneous robot cluster. Otherwise, the deployment package is refused to be loaded, and an anomaly is reported to the management layer to trigger the regeneration of control intent data.
[0045] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Acquire control intent data generated by the control layer, and perform differentiated preprocessing based on the source type of the control intent data, and output the original control intent data or a limited intermediate representation containing only capability-level abstract descriptions. The control intent data or restricted intermediate representation is subject to a three-layer progressive mandatory verification on the control side. After the verification is passed, a deployment package carrying a compiler signature or verification digest is compiled and generated. After atomic verification by the internal components of the deployment package, the deployment package is sent to the edge execution module via the communication bus. The internal components of the deployment package include task logic, version number, and security configuration data that matches the local capability list of the edge execution module. The edge execution module performs edge-side verification on the received deployment package. If the verification passes, all components in the deployment package are loaded and activated to carry out safety management and control operations for the heterogeneous robot cluster. Otherwise, the deployment package is refused to be loaded, and an anomaly is reported to the management layer to trigger the regeneration of control intent data.
[0046] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
[0047] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0048] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application.
Claims
1. A secure compilation and verification method for robot swarms, characterized in that, The method is applied to a safety management system for a heterogeneous robot cluster, including a control layer and an edge execution module. The method includes: The control intent data generated by the control layer is obtained, and differential preprocessing is performed according to the source type of the control intent data, and the original control intent data or a limited intermediate representation containing only capability level abstract description is output accordingly. The control intent data or restricted intermediate representation undergoes a three-layer progressive mandatory verification on the control and management side. Upon successful verification, a deployment package carrying a compiler signature or verification digest is compiled and generated. After atomic verification by the components within the deployment package, the deployment package is sent to the edge execution module via the communication bus. The components within the deployment package include task logic, a version number, and security configuration data matching the local capability list of the edge execution module. The three-layer progressive mandatory verification includes: The first layer of structural verification verifies whether the node or skill identifier referenced by the control intent data or restricted intermediate representation exists and is active in the capability whitelist of the edge execution module. The second layer of parameter constraint verification verifies each parameter corresponding to the node or skill identifier referenced by the control intent data or restricted intermediate representation, including whether the data type of the parameter is consistent with the data type defined in the capability whitelist, whether the parameter value is within the legal range defined in the capability whitelist, and whether the parameter enumeration value is a legal enumeration value defined in the capability whitelist. The third layer of sequence invariant verification verifies whether the action sequence corresponding to the control intent data or restricted intermediate representation violates physical security constraints based on the device security event log and preset sequence-level physical security invariants. If any layer of verification fails, the deployment package compilation is rejected, a verification failure event or structured feedback message is generated, and reported to the management layer, which then regenerates the control intent data. If the three-layer progressive mandatory verification passes, a deployment package carrying a compiler signature or verification digest is compiled and generated. The edge execution module performs edge-side verification on the received deployment package. If the verification passes, all components in the deployment package are loaded and activated to carry out safety management and control operations for the heterogeneous robot cluster. Otherwise, the deployment package is refused to be loaded, and an anomaly is reported to the management and control layer to trigger the regeneration of control intent data.
2. The secure compilation and verification method for robot swarms according to claim 1, characterized in that, Differential preprocessing is performed based on the source type of the control intent data, resulting in either the original control intent data or a limited intermediate representation containing only capability-level abstract descriptions, including: For control intent data derived from manual programming or automated tools, remain unchanged; For control intent data automatically generated by the nondeterministic strategy generator, the adapter parses the intent tags and maps each intent tag to a registered skill identifier in the capability whitelist of the edge execution module. The qualifiers in the control intent data are mapped to preset parameter levels. Based on the mapping results, a restricted intermediate representation is generated. The restricted intermediate representation includes the skill identifier, parameter level, target reference, and preconditions at the capability level, but does not contain underlying motor control commands or direct hardware operation instructions.
3. The secure compilation and verification method for robot swarms according to claim 1, characterized in that, The sequence-level physical safety invariants include: a speed limit constraint that defines the speed parameter of a single action as not exceeding the physical limit of the device; a geofence constraint that defines the target coordinates as being within the geofence polygon range that the device is allowed to operate in; a power budget constraint that defines the estimated total power consumption of the action sequence as not exceeding the device's current power consumption minus the minimum power required for a safe return; an attitude constraint that defines the device attitude changes required by the action sequence as not exceeding the safe range under the current terrain conditions; and a dangerous action cooling cycle constraint that defines high-risk actions must not reappear in the current action sequence within the minimum cooling interval after the last execution.
4. The secure compilation and verification method for robot swarms according to claim 1, characterized in that, The atomic verification of the components within the deployment package includes: verifying whether the task logic, version number, and security configuration data matching the local capability list of the edge execution module in the deployment package all exist and are compatible with each other. If any component fails verification, the deployment package is rejected and an event is generated and reported to the management layer, which then regenerates the control intent data. If all components are verified successfully, the deployment package is sent to the edge execution module via the communication bus.
5. The secure compilation and verification method for robot swarms according to claim 4, characterized in that, The security configuration data includes a set of security contracts and data valve configurations; The set of security contracts includes a security contract bound to each control interface that is allowed to be remotely invoked before the control interface of the robot device is allowed to be remotely invoked. Each security contract contains a TTL field and a security degradation action identifier field bound to the corresponding control interface, as well as an edge veto right trigger condition field bound to a single control interface or the device's global security domain. The security contract is used to perform TTL timeout monitoring of the corresponding control interface at a higher frequency than that of the task execution thread in the edge execution module after it is issued to the edge execution module and loaded and activated. It also performs veto right monitoring of the corresponding control interface at the highest priority based on the edge veto right trigger condition field. When a TTL timeout is detected and the edge hardware-level veto right is triggered, the security degradation action of the corresponding control interface is autonomously controlled according to the security degradation action identifier field without relying on network communication with the management layer. At the same time, the task execution thread is notified to interrupt or cancel the current task logic, an event record is generated and written to the local security event log and / or reported to the management layer. The data valve configuration is used to perform semantic transformation on the raw sensor data uploaded by the device through a bypass data channel independent of the communication bus after being sent to and activated by the edge execution module. This generates semantic key-value pairs and applies four-dimensional hard constraints in the data valve configuration to filter them, obtaining semantic key-value pairs that satisfy all hard constraints. The task execution thread then consumes the latest semantic key-value pairs that match the version number of the deployment package to execute the task logic. The four-dimensional hard constraints include a preset maximum field count limit, a preset maximum byte count limit, a preset maximum output frequency limit, and a preset maximum conversion delay limit for the semantic key-value pairs.
6. The secure compilation and verification method for robot swarms according to claim 1, characterized in that, The edge-side verification includes: The deployment package verification unit in the edge execution module verifies whether the compiler signature or verification digest in the issued deployment package exists, is damaged, or has been tampered with, or whether the signature algorithm is mismatched, and verifies whether the version number of the deployment package is compatible with the local capability list.
7. A secure compilation and verification device for robot swarms, characterized in that, The device is applied to a safety management system for a heterogeneous robot swarm, which includes a control layer and an edge execution module. The device includes: The differential preprocessing unit is used to acquire the control intent data generated by the control layer, and perform differential preprocessing according to the source type of the control intent data, and output the original control intent data or a limited intermediate representation containing only the capability level abstract description. The control-side verification unit performs a three-layer progressive mandatory verification on the control intent data or restricted intermediate representation. After successful verification, it compiles and generates a deployment package carrying a compiler signature or verification digest. Following atomic verification by the components within the deployment package, the deployment package is then sent to the edge execution module via the communication bus. The components within the deployment package include task logic, a version number, and security configuration data matching the local capability list of the edge execution module. The three-layer progressive mandatory verification includes: The first layer of structural verification verifies whether the node or skill identifier referenced by the control intent data or restricted intermediate representation exists and is active in the capability whitelist of the edge execution module. The second layer of parameter constraint verification verifies each parameter corresponding to the node or skill identifier referenced by the control intent data or restricted intermediate representation, including whether the data type of the parameter is consistent with the data type defined in the capability whitelist, whether the parameter value is within the legal range defined in the capability whitelist, and whether the parameter enumeration value is a legal enumeration value defined in the capability whitelist. The third layer of sequence invariant verification verifies whether the action sequence corresponding to the control intent data or restricted intermediate representation violates physical security constraints based on the device security event log and preset sequence-level physical security invariants. If any layer of verification fails, the deployment package compilation is rejected, a verification failure event or structured feedback message is generated, and reported to the management layer, which then regenerates the control intent data. If the three-layer progressive mandatory verification passes, a deployment package carrying a compiler signature or verification digest is compiled and generated. The edge-side verification unit is used by the edge execution module to perform edge-side verification on the received deployment package. If the verification is successful, all components in the deployment package are loaded and activated to carry out safety management and control operations for the heterogeneous robot cluster; otherwise, the deployment package is refused to be loaded and an anomaly is reported to the management and control layer to trigger the regeneration of control intent data.
8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.
Citation Information
Patent Citations
Multi-mode intention editing and multi-target execution compiling system based on intention protocol
CN122389844A
Robot control instruction compiling method and device and storage medium
CN122593791A
