A robot cooperative control method, a robot, a storage medium and a program product

Abstract

The application relates to the technical field of robots, and discloses a robot cooperative control method, a robot, a storage medium and a program product. The method comprises the following steps: acquiring first sensing state information of a target robot and N first relative state information of N cooperative robots relative to the target robot; encoding the first sensing state information to obtain a first encoding sequence, and respectively encoding the N first relative state information to obtain N second encoding sequences; acquiring a third encoding sequence based on the first encoding sequence, the N second encoding sequences and spatial position information of the N cooperative robots relative to the target robot; processing the third encoding sequence based on an attention mechanism to obtain first feature information, wherein the first feature information represents importance information of the N cooperative robots relative to the target robot; and determining first target action information of the target robot based on the first feature information. In this way, the flexibility of multi-robot cooperation can be improved.

Description

Technical Field

[0001] This application relates to the field of robotics technology, and in particular to a robot collaborative control method, a robot, a storage medium, and a program product. Background Technology

[0002] With the development of humanoid robot technology, robots are increasingly required to work in collaboration with multiple robots in scenarios such as logistics handling, disaster relief, and household services (e.g., multiple robots working together to move complex heavy objects, moving in formation, etc.).

[0003] Current multi-robot collaborative solutions can only be applied to scenarios with a fixed number of robots. If a robot goes offline or a new robot is added, the operation will be impossible. Furthermore, the order in which each robot acquires interaction status data from other collaborating robots must be fixed. If the order of interaction status data changes in complex scenarios, it will also lead to inaccurate data processing. In other words, current multi-robot collaborative solutions have poor flexibility. Summary of the Invention

[0004] To improve the flexibility of multi-machine collaboration, embodiments of this application provide a robot collaborative control method, a robot, a storage medium, and a program product.

[0005] In a first aspect, embodiments of this application provide a robot cooperative control method for a target robot. The method includes: acquiring first perception state information of the target robot and N first relative state information of N cooperative robots relative to the target robot, where N is a positive integer; encoding the first perception state information to obtain a first encoding sequence, and encoding the N first relative state information corresponding to the N cooperative robots to obtain N second encoding sequences corresponding to the N cooperative robots; acquiring a third encoding sequence based on the first encoding sequence, the N second encoding sequences, and the spatial position information of the N cooperative robots relative to the target robot; processing the third encoding sequence based on an attention mechanism to obtain first feature information, wherein the first feature information represents the feature information of the N cooperative robots, the feature information of the target robot, and the importance information of the N cooperative robots relative to the target robot; determining first target action information of the target robot based on the first feature information; and controlling the movement of the target robot based on the first target action information.

[0006] In the above scheme, the state data of the main robot and other collaborative robots are encoded to obtain an encoded sequence. The input sequence is then further processed using an attention mechanism. It can be understood that the attention mechanism has no limit on the length of the input sequence. Therefore, it can process the data of any number of collaborative robots, thereby adapting to the dynamic changes in the number of collaborative robots and improving the application flexibility and scenario adaptability of the scheme.

[0007] Secondly, based on the disordered nature of the attention mechanism, this solution does not need to rely on a fixed data input order or robot identification (ID) number when processing the state information of each collaborative robot. It is completely unaffected by the adjustment of data input order or the switching of robot ID numbers. Even in complex dynamic environments where robot formations are dynamically reorganized and collaborative members change frequently, it can ensure the stable output of the control strategy, further improving the application flexibility of the solution.

[0008] Furthermore, this solution processes the third encoded sequence through an attention mechanism, which can automatically quantify the "interaction importance" of the surrounding collaborative robots, and obtain the robot's next action information based on the features that incorporate the interaction importance, thereby improving the accuracy of action determination.

[0009] In one possible implementation of the first aspect, the first relative state information of the collaborative robot relative to the target robot includes: relative pose information and relative velocity information of the collaborative robot relative to the target robot.

[0010] In one possible implementation of the first aspect, the spatial position information of the N collaborative robots relative to the target robot includes the distance information of each of the N collaborative robots relative to the target robot and the topological position information.

[0011] In one possible implementation of the first aspect, obtaining a third encoding sequence based on a first encoding sequence, N second encoding sequences, and the spatial position information of N collaborative robots relative to the target robot includes: determining the third encoding sequence based on the first encoding sequence, N second encoding sequences, the spatial position information of N collaborative robots relative to the target robot, and the role identification information of N collaborative robots.

[0012] In one possible implementation of the first aspect, a third encoding sequence is obtained based on a first encoding sequence, N second encoding sequences, and N spatial position information of collaborative robots relative to a target robot; including: generating N sets of spatial position embedding vectors based on the spatial position information of the N collaborative robots relative to the target robot; superimposing the N sets of spatial position embedding vectors with the corresponding N second encoding sequences to obtain N fourth encoding sequences; and obtaining the third encoding sequence based on the first encoding sequence and the N fourth encoding sequences.

[0013] In one possible implementation of the first aspect, processing the third encoded sequence based on an attention mechanism to obtain first feature information includes: processing the third encoded sequence based on an attention model to obtain the first feature information and an attention weight matrix corresponding to N collaborative robots, wherein the attention weight matrix includes the attention weight value of each collaborative robot; the method further includes: determining M collaborative robots based on the attention weight matrix corresponding to the N collaborative robots, wherein the attention weight values ​​of the M collaborative robots are greater than a first threshold, and M is a positive integer less than N; obtaining the second perceptual state information of the target robot and M second relative state information of the M collaborative robots relative to the target robot; and determining the second target action information of the target robot based on the second perceptual state information of the target robot and the M second relative state information of the M collaborative robots relative to the target robot.

[0014] In one possible implementation of the first aspect, N collaborative robots are located within a preset range of the target robot.

[0015] In one possible implementation of the first aspect, the first perceived state information includes state information within a preset time period prior to the current moment; the state information includes at least one of the following: joint position information, velocity information, drive command information, and sensor data information.

[0016] In one possible implementation of the first aspect, determining the target motion information of the target robot based on the first feature information includes: processing the first feature information and the first perception state information based on the motion decision network, and outputting the target motion information of the target robot, wherein the target motion information includes the joint position information and / or torque information of the target robot.

[0017] In a second aspect, embodiments of this application provide a robot, including: one or more processors; one or more memories; the one or more memories storing one or more programs, which, when executed by the processors, cause the robot to perform the method of any one of the first aspects.

[0018] Thirdly, embodiments of this application provide a robot collaborative control system, including a target robot and multiple collaborative robots, the multiple collaborative robots including N collaborative robots, and the target robot is used to perform the method of any one of the first aspects.

[0019] Fourthly, embodiments of this application provide a computer-readable storage medium storing a program or instructions that, when executed by an electronic device, implement the method as described in any of the first aspects.

[0020] Fifthly, embodiments of this application provide a computer program product including instructions that, when executed, cause the method of any one of the first aspects to be implemented. Attached Figure Description

[0021] Figure 1 According to an embodiment of this application, a flowchart of a robot cooperative control method is illustrated.

[0022] Figure 2 According to an embodiment of this application, a schematic diagram of a scenario in which multiple robots cooperate to transport materials is shown.

[0023] Figure 3 According to an embodiment of this application, a flowchart of a collaborative control method is illustrated.

[0024] Figure 4 According to an embodiment of this application, a schematic diagram of the structure of a robot is shown. Detailed Implementation

[0025] This application provides a robot collaborative control method, a robot, a storage medium, and a program product.

[0026] As mentioned earlier, with the development of humanoid robot technology, robots are increasingly required to work in collaboration with multiple robots in scenarios such as logistics handling, disaster relief, and household services (e.g., multiple robots working together to move complex heavy objects, moving in formation, etc.).

[0027] In some embodiments, robots typically deploy artificial intelligence models (e.g., policy networks) to perform collaborative tasks. However, multilayer perceptrons (MLPs) are commonly used as policy networks. It's understood that the input layer dimension of an MLP is fixed, meaning the dimension of the input data is fixed. Therefore, during collaborative operations, the dimension of data that can be received from the robots is fixed, resulting in a fixed number of robots (e.g., two) working collaboratively. If a new robot is added or a robot goes offline, the dimension of the input data changes. This can lead to a mismatch between the dimension of the input layer and the dimension of the input data, preventing the robot from processing data and causing task execution failure.

[0028] Furthermore, in multi-robot collaborative systems, the order in which a robot receives the state arrays of other collaborating robots is fixed. For example, if the order of the collaborative robots' state arrays is reversed in a complex environment, it will cause the policy network deployed in the robot to output completely different or disordered joint movements, reducing the control stability and robustness during complex dynamic interactions between multiple robots.

[0029] In summary, the above-mentioned collaborative control schemes have poor application flexibility.

[0030] To address the aforementioned issues, this application provides a robot cooperative control method for a target robot. The method includes: acquiring first perception state information of the target robot and N first relative state information of N cooperative robots relative to the target robot, where N is a positive integer; encoding the first perception state information to obtain a first encoding sequence, and encoding the N first relative state information corresponding to the N cooperative robots to obtain N second encoding sequences corresponding to the N cooperative robots; acquiring a third encoding sequence based on the first encoding sequence, the N second encoding sequences, and the spatial position information of the N cooperative robots relative to the target robot; processing the third encoding sequence based on an attention mechanism to obtain first feature information, where the first feature information represents the feature information of the N cooperative robots, the feature information of the target robot, and the importance information of the N cooperative robots relative to the target robot; determining first target action information of the target robot based on the first feature information; and controlling the movement of the target robot based on the first target action information.

[0031] In the above scheme, the state data of the main robot and other collaborative robots are encoded to obtain an encoded sequence. The input sequence is then further processed using an attention mechanism. It can be understood that the attention mechanism has no limit on the length of the input sequence. Therefore, it can process the data of any number of collaborative robots, thereby adapting to the dynamic changes in the number of collaborative robots and improving the application flexibility and scenario adaptability of the scheme.

[0032] Secondly, based on the disordered nature of the attention mechanism, this solution does not need to rely on a fixed data input order or robot identification (ID) number when processing the state information of each collaborative robot. It is completely unaffected by the adjustment of data input order or the switching of robot ID numbers. Even in complex dynamic environments where robot formations are dynamically reorganized and collaborative members change frequently, it can ensure the stable output of the control strategy, further improving the application flexibility of the solution.

[0033] Furthermore, this solution processes the third encoded sequence through an attention mechanism, which can automatically quantify the "interaction importance" of the surrounding collaborative robots, and obtain the robot's next action information based on the features that incorporate the interaction importance, thereby improving the accuracy of action determination.

[0034] The robot cooperative control method provided in the embodiments of this application is described below.

[0035] Figure 1 The diagram illustrates a robot collaborative control method, which can be executed by a robot and may include:

[0036] 101: Obtain the first perception state information of the target robot and the N first relative state information of N collaborative robots relative to the target robot, where N is a positive integer.

[0037] In some embodiments, the target robot and the N collaborating robots can be robots that jointly perform the same task. For example, five robots can be set up to perform a handling task. All robots jointly performing the same task can execute the collaborative control method provided in the embodiments of this application.

[0038] In some embodiments, the N collaborative robots can be robots performing the same task that are located within a preset range of the target robot. The preset range can be the detection range or communication range of the target robot's radar sensor. That is, the N collaborative robots can be robots located within the detection range or communication range of the target robot's radar sensor. For example, if 10 robots perform the same handling task, but only 4 robots are within the detection range of the target robot's radar sensor, then the N robots can refer to those 4 robots.

[0039] In some embodiments, the preset range can be other values ​​set according to requirements, and this application embodiment does not limit it.

[0040] In some embodiments, the first perceived state information of the target robot includes state information within a preset time period (e.g., 2 seconds) prior to the current moment; the state information includes at least one of the following: joint position information, velocity information, drive command information, and sensor data information.

[0041] In some embodiments, the first relative state information of the collaborative robot relative to the target robot includes: relative pose information and relative velocity information of the collaborative robot relative to the target robot.

[0042] In some embodiments, the target robot can establish communication with each collaborating robot and receive its own position information, such as position coordinates, sent by each collaborating robot based on the communication.

[0043] In some embodiments, after receiving the position information sent by each collaborating robot, the target robot can determine the relative pose and relative velocity information of each collaborating robot relative to the target robot based on its own position information.

[0044] 102: Encode the first perception state information to obtain the first encoding sequence, and encode the N first relative state information corresponding to the N collaborative robots to obtain the N second encoding sequences corresponding to the N collaborative robots.

[0045] In some embodiments, the first perceived state information can be mapped to a first encoded sequence of a unified dimension through an encoding module. The first encoded sequence can also be called an Ego Token.

[0046] In some embodiments, the encoding module can encode the N first relative state information corresponding to the N collaborative robots respectively, and map them into N second encoding sequences, which can also be called agent tokens.

[0047] 103: Based on the first encoding sequence, N second encoding sequences, and N spatial position information of the collaborative robots relative to the target robot, obtain the third encoding sequence.

[0048] In some embodiments, obtaining the third encoding sequence based on the first encoding sequence, N second encoding sequences, and the spatial position information of the N collaborative robots relative to the target robot may include: generating N sets of spatial position embedding vectors based on the spatial position information of the N collaborative robots relative to the target robot; superimposing the N sets of spatial position embedding vectors with the corresponding N second encoding sequences to obtain N fourth encoding sequences; and obtaining the third encoding sequence based on the first encoding sequence and the N fourth encoding sequences. That is, the N sets of spatial position embedding vectors can be first superimposed with the corresponding N second encoding sequences to obtain N fourth encoding sequences, and then the fourth encoding sequences can be merged with the first encoding sequence along the length direction to obtain the third encoding sequence.

[0049] It is understandable that the dimensions of the second encoding sequence and the spatial embedding vector are consistent. Therefore, the dimensions can be superimposed to obtain a new fourth encoding sequence.

[0050] For example, taking four collaborative robots as an example, the second encoding sequences corresponding to the four collaborative robots are AgentToken1, AgentToken2, AgentToken3, and AgentToken4, respectively. Based on the spatial position information of the four collaborative robots relative to the target robot, four sets of spatial position embedding vectors are generated, namely PosEmb1, PosEmb2, PosEmb3, and PosEmb4. Each set of spatial position embedding vectors is superimposed dimension-wise with the second encoding sequence of the corresponding collaborative robot to obtain four fourth encoding sequences: AgentToken1+PosEmb1, AgentToken2+PosEmb2, AgentToken3+PosEmb3, and AgentToken4+PosEmb4. Subsequently, the first encoding sequence (Ego Token) corresponding to the target robot is concatenated and fused with the above four fourth encoding sequences in the sequence length direction, resulting in the third encoding sequence {Ego Token, Agent Token1+PosEmb1, Agent Token2+PosEmb2, Agent Token3+PosEmb3, AgentToken4+PosEmb4}.

[0051] In some embodiments, the first encoding sequence and N second encoding sequences can be merged along the length direction to obtain a merged encoding sequence. Based on the spatial position information of the N collaborative robots relative to the target robot, N sets of spatial position embedding vectors are generated; the N sets of spatial position embedding vectors are then superimposed with the corresponding N second encoding sequences in the merged encoding sequence to obtain a third encoding sequence.

[0052] In some embodiments, the spatial position information of the N collaborative robots relative to the target robot can be the distance information of each of the N collaborative robots relative to the target robot or the topology connection map information.

[0053] In some embodiments, obtaining a third encoding sequence based on a first encoding sequence, N second encoding sequences, and the spatial position information of the N collaborative robots relative to the target robot may include: determining the third encoding sequence based on the first encoding sequence, N second encoding sequences, the spatial position information of the N collaborative robots relative to the target robot, and the role identification information of the N collaborative robots. That is, the role identification information of the N collaborative robots can be added to the encoding sequence.

[0054] For example, after merging the first and second encoding sequences along their length (e.g., forming an N+1 length sequence), the role identification information of the N robots can be added to the corresponding N second encoding sequences to form the final third encoding sequence. The role identification can be based on the role of each collaborative robot in the current task. For example, the role identification could be "close collaborator," "follower in formation," or "obstacle avoidance required."

[0055] In some embodiments, the role identifiers of all robots can be pre-stored in each robot, meaning the target robot can determine the role identifier information of the N collaborative robots based on the stored role identifier information of each robot. In some embodiments, each robot can also broadcast its role identifier information to other robots via communication, meaning the target robot can receive role identifier information sent by each of the N collaborative robots based on communication with the N collaborative robots.

[0056] 104: The third encoding sequence is processed based on the attention mechanism to obtain the first feature information. The first feature information represents the feature information of N collaborative robots, the feature information of the target robot, and the importance information of the N collaborative robots relative to the target robot.

[0057] In some embodiments, processing the third encoded sequence based on an attention mechanism to obtain first feature information may include: processing the third encoded sequence based on an attention model to obtain the first feature information and an attention weight matrix corresponding to N collaborative robots. The attention weight matrix includes the attention weight value of each collaborative robot relative to the target robot. It can be understood that the attention weight matrix may include the attention weight values ​​of each robot relative to all other robots.

[0058] In some embodiments, the attention model can be an attention model based on the Transformer architecture.

[0059] In some embodiments, the self-attention model can use the first encoding sequence (Ego Token) as the query vector, the second encoding sequence superimposed with the spatial embedding vector, or the second encoding sequence superimposed with the spatial embedding vector and the added role identification information as the key and value, to calculate the attention score and output the first feature information and the attention weight matrix.

[0060] The first feature information is a context feature vector that integrates global collaborative importance ranking, such as the feature information of N collaborative robots, the feature information of the target robot, and the importance information of the N collaborative robots relative to the target robot.

[0061] 105: Determine the first target action information of the target robot based on the first feature information.

[0062] In some embodiments, determining the target motion information of the target robot based on the first feature information may include: inputting the first feature information and the first perception state information into a motion decision network, processing the first feature information and the first perception information based on the motion decision network, and outputting the target motion information of the target robot. The target motion information may include the joint position information and / or torque information of the target robot.

[0063] 106: Control the movement of the target robot based on the first target's motion information.

[0064] In some embodiments, after acquiring the first target motion information, the target robot's movement can be controlled based on the target motion information.

[0065] In some embodiments, the robot can dynamically determine, based on the attention weight matrix, whether to receive or process only collaborative robot data frames whose weights are higher than a first threshold in the next frame.

[0066] For example, based on the attention weight matrix corresponding to N collaborative robots, M collaborative robots can be determined, where the attention weight values ​​of the M collaborative robots are greater than a first threshold.

[0067] After determining the M collaborative robots, in the next frame of data processing, the second perception state information of the target robot and the M second relative state information of the M collaborative robots relative to the target robot can be obtained; based on the second perception state information of the target robot and the M second relative state information of the M collaborative robots relative to the target robot, the second target action information of the target robot is determined, where M is a positive integer less than N.

[0068] For example, taking four collaborative robots (collaborative robots R1-R4) as an example, the weight values ​​of collaborative robots R1-R4 relative to the target robot are [0.09, 0.54, 0.29, 0.08], respectively. Taking a first threshold of 0.20 as an example, comparing the weight values ​​of collaborative robots R1-R4 relative to the target robot with the first threshold, the collaborative robots with weight values ​​greater than the first threshold are collaborative robots R2 and R3. Therefore, collaborative robots R2 and R3 can be used as the objects for the next interaction. For example, when receiving or processing data frames from collaborative robots in the next frame (e.g., location information), only data frames from collaborative robots R2 and R3 can be received or processed, while data frames from collaborative robots R1 and R4 can be ignored.

[0069] In the above scheme, the state data of the main robot and other collaborative robots are encoded to obtain the encoded sequence, and the input sequence is further processed by the attention mechanism. It can be understood that the attention mechanism (Transformer architecture) has no limit on the length of the input sequence. Therefore, it can process the data of any number of collaborative robots, thereby adapting to the dynamic changes in the number of collaborative robots and improving the application flexibility and scenario adaptability of the scheme.

[0070] Secondly, based on the disordered nature of the attention mechanism, this solution does not need to rely on a fixed data input order or robot ID number when processing the state information of each collaborative robot. It is completely unaffected by the adjustment of the data input order or the switching of robot ID numbers. Even in complex dynamic environments where robot formations are dynamically reorganized and collaborative members change frequently, it can ensure the stable output of the control strategy, further improving the application flexibility of the solution.

[0071] Furthermore, this solution processes the third encoded sequence through an attention mechanism, which can automatically quantify the "interaction importance" of the surrounding collaborative robots. Based on the features that integrate the interaction importance, the solution can obtain the robot's next action information, thereby improving the accuracy of action determination.

[0072] Furthermore, in multi-robot physical environments, not all robots have equal interaction priorities (e.g., robots at greater distances have minimal impact on the current collision risk). Existing methods lack dynamic assessment of the importance of collaborating objects and typically process data from all robots, leading to severe resource waste and communication latency in real-world deployments with limited edge computing power and wireless communication bandwidth.

[0073] The solution provided in this application can automatically quantify the "interaction importance" of surrounding robots. Under the bottleneck of local area network communication in multi-robot formation, attention weights can be used to actively discard information of collaborative robots with lower importance, and prioritize data exchange of high-weight nodes, realizing intelligent decoupling of physical interaction and computing resources, reducing resource waste and communication latency.

[0074] This application provides a robot collaborative control system, including a target robot and multiple collaborative robots, wherein the multiple collaborative robots include N collaborative robots, and the target robot is used to execute the collaborative control method provided in this application.

[0075] In some embodiments, the total number of target robots and collaborative robots may be greater than or equal to three.

[0076] In some embodiments, each robot in the robot collaborative control system can be deployed with an independent onboard computing unit (to achieve decentralized execution). The onboard computing unit can be used to execute the collaborative control method provided in the embodiments of this application. All robots in the cluster broadcast their global pose (e.g., the aforementioned position information) and basic observation information at a set frequency (e.g., 50Hz) through a local area network (such as the LCM communication protocol).

[0077] In some embodiments, the airborne unit may include an encoding module, an attention module, and an action decoder.

[0078] The encoding module is used to acquire the first perception state information of the target robot and the N first relative state information of the N collaborative robots relative to the target robot, encode the first perception state information to obtain a first encoding sequence, and encode the N first relative state information corresponding to the N collaborative robots to obtain N second encoding sequences corresponding to the N collaborative robots. Based on the first encoding sequence, the N second encoding sequences and the spatial position information of the N collaborative robots relative to the target robot, a third encoding sequence is acquired.

[0079] It is also used to acquire the second perceptual state information of the target robot and the M second relative state information of the M collaborative robots relative to the target robot; and to determine the second target action information of the target robot based on the second perceptual state information of the target robot and the M second relative state information of the M collaborative robots relative to the target robot.

[0080] The attention module is used to process the third encoded sequence based on the attention mechanism to obtain the first feature information. The first feature information represents the feature information of N collaborative robots, the feature information of the target robot, and the importance information of the N collaborative robots relative to the target robot.

[0081] The motion decoder is used to determine the first target motion information of the target robot based on the first feature information, and to control the movement of the target robot based on the first target motion information.

[0082] In some embodiments, each robot can encode based on the encoding module to obtain a third encoding sequence. In some embodiments, a master robot can also encode based on the encoding module and distribute the encoded third encoding sequence to the other robots.

[0083] In some embodiments, the network model including an encoding module, an attention module, and an action decoder provided in this application can be trained on a physical simulation platform and then deployed to the onboard computing unit of each robot after training.

[0084] For example, multi-robot collaborative tasks can be constructed in a physical simulation platform (such as IsaacLab), such as three robots collaboratively transporting a long, heavy object. When constructing the observation space, a dynamic environment is simulated, and 1 to 5 collaborative robots are randomly generated in different training rounds. During training, the ontological observation data (e.g., first-perception state information) and the received external broadcast poses (e.g., the collaborative robot's position information) are converted into local relative coordinates. A 64-dimensional token sequence is generated through a 1D-CNN layer with shared weights (i.e., the encoding module), and then fed into a Transformer module (i.e., the attention module) containing 3 layers and 4 attention heads (Num_heads=4) to complete contextual feature aggregation for training the network model. Regarding the reward function, in addition to setting a conventional single-unit balancing reward, a geometric bias penalty term based on a multi-node interaction graph is introduced to guide the model to learn to maintain relative topological stability among the multi-robots using attention features. The Proximal Policy Optimization (PPO) algorithm is used to iteratively update the weights of the network model to complete the training.

[0085] Figure 2 The diagram illustrates a scenario of multiple robots working together to move a long, heavy object. Taking the example of a target robot and four collaborating robots working together to move a long, heavy object, it visually demonstrates the multi-robot interactive perception process based on an attention mechanism.

[0086] exist Figure 2 In the scenario shown, the target robot, as the core node of the collaborative task, dynamically evaluates the importance of each collaborating robot using attention weights. Meanwhile, Figure 2 The spatial position of each robot is marked by a bounding box, and the relative pose information of the collaborative robot relative to the target robot (such as relative coordinates of +2.1m and -0.4m) is displayed.

[0087] Figure 3 A flowchart illustrating a collaborative control method is shown. For example... Figure 3 As shown, taking a five-machine collaborative scenario consisting of target robot R5 and collaborative robots R1-R4 as an example, the process may include:

[0088] Step 1: The target robot acquires initial perception state information through sensing devices such as cameras and LiDAR, and receives position information from the collaborative robot to obtain relative state information data. The data is then tokenized and encoded, converting different types of raw data into coded data (tokens) of a unified dimension, such as the aforementioned first and second encoding sequences.

[0089] Step 2: A spatial location identifier (e.g., a spatial location vector) is added to each second encoded sequence through positional encoding to form a third encoded sequence. This third encoded sequence is then input into the Transformer network module to obtain the output first feature information and attention weight matrix. The Transformer network module includes a Layer Normalization layer, a Multi-Head Attention (MHA) network, a Residual network, and a FeedForward Network (FFN). The Transformer network module uses the target robot token (first encoded sequence) as the query vector and the tokens of each collaborating robot (second encoded sequence) as the key and value. After linear projection and scaled dot product attention calculation, it outputs an attention weight matrix and the first feature information. This attention weight matrix represents the distribution of the importance of interactions between the target robot and each collaborating robot.

[0090] Step 3: Input the first feature information into the attention aggregation and motion control strategy module. The attention aggregation and motion control strategy module includes a precise trajectory strategy network. Through the precise trajectory strategy network, the target robot's target motion information, such as joint torque and position information, can be generated based on the cooperative feature vector (i.e., the first feature information). Based on the target motion information, the target robot is driven to perform cooperative tasks.

[0091] It is understood that, based on the method provided in this application embodiment, if a new robot joins the network and starts broadcasting data while a robot is performing a handling task, the main control program does not need to be restarted. The encoding module can automatically encode the new broadcast data into a new Agent Token and append it to the end of the current sequence. The Transformer module evaluates the spatial distance and interaction intent of the newly joined robot in real time, outputs contextual feature information and attention weight matrix, thereby enabling subsequent control of the robot's actions. In this way, seamless collaborative operation is achieved under dynamic changes in the number of robots in the physical space.

[0092] Figure 4 A schematic diagram of the structure of a robot is shown. Figure 4In the illustrated embodiment, robot 100 may include one or more processors 101, system control logic 102 connected to at least one of the processors 101, system memory 103 connected to system control logic 102, non-volatile memory (NVM) 104 connected to system control logic 102, and network interface 106 connected to system control logic 102.

[0093] In some embodiments, processor 101 may include one or more single-core or multi-core processors. In some embodiments, processor 101 may include any combination of general-purpose processors and special-purpose processors (e.g., graphics processors, application processors, baseband processors, etc.). In embodiments where robot 100 employs an Evolved Node B (eNB) or Radio Access Network (RAN) controller, processor 101 may be configured to perform various corresponding embodiments. For example, processor 101 may be used in the cooperative control method provided in embodiments of this application.

[0094] In some embodiments, system control logic 102 may include any suitable interface controller to provide any suitable interface to at least one suitable device or component in processor 101 that communicates with system control logic 102.

[0095] In some embodiments, system control logic 102 may include one or more memory controllers to provide an interface to system memory 103. System memory 103 may be used to load and store vector data and / or instructions.

[0096] In some embodiments, the system memory 103 of the robot 100 may include any suitable volatile memory, such as suitable dynamic random access memory (DRAM).

[0097] NVM memory 104 may include one or more tangible, non-transitory computer-readable media for storing data and / or instructions. In some embodiments, NVM memory 104 may include any suitable non-volatile memory such as flash memory and / or any suitable non-volatile storage device, such as at least one of a hard disk drive (HDD), a compact disc (CD) drive, and a digital versatile disc (DVD) drive.

[0098] NVM memory 104 may include a portion of the storage resources on the device on which robot 100 is mounted, or it may be accessible by the device, but is not necessarily part of the device. For example, NVM memory 104 may be accessed over a network via network interface 106.

[0099] Specifically, system memory 103 and NVM memory 104 may each include a temporary copy and a permanent copy of instruction 105. Instruction 105 may include instructions that, when executed by at least one of processors 101, cause robot 100 to implement the vector data query method as described in embodiments of this application. In some embodiments, instruction 105, hardware, firmware, and / or its software components may additionally / alternatively reside in system control logic 102, network interface 106, and / or processor 101.

[0100] Network interface 106 may include a transceiver for providing a radio interface to robot 100, thereby enabling communication with any other suitable devices (such as front-end modules, antennas, etc.) via one or more networks. In some embodiments, network interface 106 may be integrated into other components of robot 100.

[0101] In some embodiments, at least one of the processors 101 may be packaged together with the logic of one or more controllers for system control logic 102 to form a system in a package (SiP). In some embodiments, at least one of the processors 101 may be integrated on the same die with the logic of one or more controllers for system control logic 102 to form a system on a chip (SoC).

[0102] The robot 100 may further include an input / output (I / O) device 107. The I / O device 107 may include a user interface that enables a user to interact with the robot 100; the peripheral component interface is designed so that peripheral components can also interact with the robot 100.

[0103] Understandable Figure 4 The illustrated structure does not constitute a specific limitation on robot 100. In other embodiments of this application, robot 100 may include more or fewer components than illustrated, or combine some components, or separate some components, or have different component arrangements. The illustrated components may be implemented by hardware or software, or a combination of software and hardware.

[0104] This application provides a chip, including: one or more processors; one or more memories; the one or more memories storing one or more programs, which, when executed by the processors, cause the chip to execute the robot control method provided in this application.

[0105] This application provides a robot, including: one or more processors; one or more memories; the one or more memories store one or more programs, which, when executed by the processors, cause the robot to perform the robot cooperative control method provided in this application.

[0106] This application provides a computer-readable storage medium storing a program or instructions. When the program or instructions are executed by an electronic device, the robot collaborative control method provided in this application is implemented.

[0107] This application provides a computer program product, including instructions, which, when executed, enable the robot cooperative control method provided in this application to be implemented.

[0108] This application provides an electronic device, including: one or more processors; one or more memories; the one or more memories storing one or more programs, which, when executed by the processors, cause the electronic device to perform the robot cooperative control method provided in this application.

[0109] In some embodiments, the electronic device may be the target robot mentioned above or any collaborative robot. In some embodiments, the electronic device may also be a control device for controlling the target robot and the collaborative robot.

[0110] Various embodiments of the mechanisms disclosed in this application can be implemented in hardware, software, firmware, or combinations of these implementation methods. Embodiments of this application can be implemented as computer programs or program code executable on a programmable system, the programmable system including at least one processor, a storage system (including volatile and non-volatile memory and / or storage elements), at least one input device, and at least one output device.

[0111] Program code can be applied to input instructions to execute the functions described in this application and generate output information. The output information can be applied to one or more output devices in a known manner. For the purposes of this application, the processing system includes any system having a processor such as, for example, a digital signal processor, a microcontroller, an application-specific integrated circuit, or a microprocessor.

[0112] The program code can be implemented using a high-level procedural language or an object-oriented programming language to communicate with the processing system. Assembly language or machine language can also be used when needed. In fact, the mechanisms described in this application are not limited to any particular programming language. In either case, the language can be a compiled language or an interpreted language.

[0113] In some cases, the disclosed embodiments may be implemented in hardware, firmware, software, or any combination thereof. The disclosed embodiments may also be implemented as instructions carried or stored on or on one or more temporary or non-temporary machine-readable (e.g., computer-readable) storage media, which may be read and executed by one or more processors. For example, the instructions may be distributed via a network or through other computer-readable media. Therefore, machine-readable media may include any mechanism for storing or transmitting information in a machine-readable (e.g., computer-readable) form, including but not limited to floppy disks, optical disks, optical discs, read-only memory, magneto-optical disks, random access memory, erasable programmable read-only memory, electrically erasable programmable read-only memory, magnetic cards or optical cards, flash memory, or tangible machine-readable storage for transmitting information (e.g., carrier waves, infrared signals, digital signals, etc.) using the Internet in the form of electrical, optical, acoustic, or other propagation signals. Therefore, machine-readable media include any type of machine-readable medium suitable for storing or transmitting electronic instructions or information in a machine-readable (e.g., computer-readable) form.

[0114] In the accompanying drawings, some structural or methodological features may be shown in a specific arrangement and / or order. However, it should be understood that such a specific arrangement and / or order may not be necessary. Rather, in some embodiments, these features may be arranged in a manner and / or order different from that shown in the illustrative drawings. Furthermore, the inclusion of structural or methodological features in a particular figure does not imply that such features are required in all embodiments, and in some embodiments, these features may be omitted or may be combined with other features.

[0115] It should be noted that all units / modules mentioned in the device embodiments of this application are logical units / modules. Physically, a logical unit / module can be a physical unit / module, a part of a physical unit / module, or a combination of multiple physical units / modules. The physical implementation of these logical units / modules themselves is not the most important factor; the combination of functions implemented by these logical units / modules is the key to solving the technical problems proposed in this application. Furthermore, to highlight the innovative aspects of this application, the above-described device embodiments of this application have not introduced units / modules that are not closely related to solving the technical problems proposed in this application. This does not mean that the above-described device embodiments do not contain other units / modules.

[0116] It should be noted that in the examples and description of this patent, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0117] Although this application has been illustrated and described with reference to certain preferred embodiments thereof, those skilled in the art should understand that various changes in form and detail may be made thereto without departing from the spirit and scope of this application.

Claims

1. A robot cooperative control method, characterized in that, For a target robot, the method includes: Acquire the first perception state information of the target robot and the N first relative state information of N collaborative robots relative to the target robot, where N is a positive integer; The first perceived state information is encoded to obtain a first encoded sequence, and the N first relative state information corresponding to the N collaborative robots are encoded respectively to obtain N second encoded sequences corresponding to the N collaborative robots; Based on the first encoding sequence, the N second encoding sequences, and the spatial position information of the N collaborative robots relative to the target robot, a third encoding sequence is obtained; The third encoded sequence is processed based on an attention mechanism to obtain first feature information. The first feature information represents the feature information of the N collaborative robots, the feature information of the target robot, and the importance information of the N collaborative robots relative to the target robot. Based on the first feature information, the first target action information of the target robot is determined; The target robot is controlled to move based on the first target motion information; The step of obtaining the third encoding sequence based on the first encoding sequence, the N second encoding sequences, and the spatial position information of the N collaborative robots relative to the target robot includes: Based on the first encoding sequence, the N second encoding sequences, the spatial position information of the N collaborative robots relative to the target robot, and the role identification information of the N collaborative robots, the third encoding sequence is determined, wherein the role identification information of the N collaborative robots is based on the role settings of the N collaborative robots in the current task.

2. The robot cooperative control method according to claim 1, characterized in that, The first relative state information of the collaborative robot relative to the target robot includes: The relative pose and relative velocity information of the collaborative robot relative to the target robot.

3. The robot cooperative control method according to claim 1, characterized in that, The spatial position information of the N collaborative robots relative to the target robot includes the distance information and topological position information of each of the N collaborative robots relative to the target robot.

4. The robot cooperative control method according to any one of claims 1-3, characterized in that, The step of determining the third encoding sequence based on the first encoding sequence, the N second encoding sequences, the spatial position information of the N collaborative robots relative to the target robot, and the role identification information of the N collaborative robots includes: Based on the spatial position information of the N collaborative robots relative to the target robot, N sets of spatial position embedding vectors are generated; The N sets of spatial location embedding vectors are superimposed on the corresponding N second coding sequences to obtain N fourth coding sequences; The third encoding sequence is obtained based on the first encoding sequence, the N fourth encoding sequences, and the role identification information of the N collaborative robots.

5. The robot cooperative control method according to any one of claims 1-3, characterized in that, The process of processing the third encoded sequence based on the attention mechanism to obtain the first feature information includes: The third encoded sequence is processed based on the attention model to obtain the first feature information and the attention weight matrix corresponding to the N collaborative robots. The attention weight matrix includes the attention weight value of each collaborative robot. The method further includes: Based on the attention weight matrix corresponding to the N collaborative robots, M collaborative robots are determined, wherein the attention weight values ​​corresponding to the M collaborative robots are greater than the first threshold, and M is a positive integer less than N; Acquire the second perception state information of the target robot and the M second relative state information of the M collaborative robots relative to the target robot; The second target action information of the target robot is determined based on the second perception state information of the target robot and the M second relative state information of the M cooperative robots relative to the target robot.

6. The robot cooperative control method according to any one of claims 1-3, characterized in that, The N collaborative robots are located within a preset range of the target robot.

7. The robot cooperative control method according to any one of claims 1-3, characterized in that, The first sensed state information includes state information within a preset time period prior to the current moment; The status information includes at least one of the following: joint position information, speed information, drive command information, and sensor data information.

8. The robot cooperative control method according to any one of claims 1-3, characterized in that, Determining the first target action information of the target robot based on the first feature information includes: The first feature information and the first perception state information are processed based on the action decision network to output the first target action information of the target robot. The first target action information includes the joint position information and / or torque information of the target robot.

9. A robot, characterized in that, include: One or more processors; One or more memories; the one or more memories storing one or more programs that, when executed by the processor, cause the robot to perform the robot cooperative control method of any one of claims 1-8.

10. A robot cooperative control system, characterized in that, It includes a target robot and multiple collaborative robots, the multiple collaborative robots comprising N collaborative robots, the target robot being used to execute the robot collaborative control method of any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program or instructions that, when executed by an electronic device, implement the robot cooperative control method as described in any one of claims 1 to 8.

12. A computer program product, characterized in that, The instructions include, when executed, causing the robot cooperative control method of any one of claims 1-8 to be implemented.

Images