Periodic Reporting for period 2 - CONVINCE (CONtext-aware VerifIable dyNamiC dEliberation)
Période du rapport: 2023-10-01 au 2024-09-30
There is growing expectation for the deployment of robots that can work for extended periods of time in unstructured environments without human intervention. To achieve this goal, robots must be programmed to perform useful tasks and handle unexpected situations by activating contingency plans to remain operational and avoid catastrophic failures.
The key contribution of CONVINCE is to develop cognitive deliberation capabilities that ensure robot operation over extended periods of time without human intervention. CONVINCE provides tools that allow automatic verification of such capabilities at design and run-time. The project will benefit the software developers of the robotic community by developing an integrated toolchain for designing application-specific deliberation systems. These systems can determine the necessary behaviors of the robot to fulfill a given task, taking into account the context in which the robot operates, and the experience gained during previous executions of the same task. The toolchain also allows for the deployment and configuration of the components that are required to execute these behaviors and automate the analysis of behaviors to ensure that they are safe and secure.
The development of CONVINCE is guided by three real-world scenarios, each posing unique challenges for robotic deliberations, which will be used to validate the results of the project.
• Robot Vacuum Cleaner (UC1): a robot that navigates in diverse home environments and should learn from its experience and avoid getting stuck.
• Assembly Robot (UC2): a robot that performs a complex task of assembling parts based on their shape, and should cope with occlusions, difficult lighting conditions, and missing or restrained parts.
• Robotic Museum Guide (UC3): a humanoid robot that guides visitors inside a museum and should interact with human and handle problematic situations that may arise from unpredictable human behavior.
In CONVINCE, a user-centered design approach is followed, where user needs are assessed to establish a set of high-level requirements. These requirements are then scrutinized concerning legal and ethical regulations, such as GDPR, Ethical AI, and the EU Artificial Intelligence Act, and connected to technical KPIs of Use Cases. Additionally, input from developers of robotic software, who are the end-users of the toolchain, is also taken into account and considered in CONVINCE to increase the success of the project and ensure the outputs meet the needs of the users.
The key contribution of CONVINCE is to develop cognitive deliberation capabilities that ensure robot operation over extended periods of time without human intervention. CONVINCE provides tools that allow automatic verification of such capabilities at design and run-time. The project will benefit the software developers of the robotic community by developing an integrated toolchain for designing application-specific deliberation systems. These systems can determine the necessary behaviors of the robot to fulfill a given task, taking into account the context in which the robot operates, and the experience gained during previous executions of the same task. The toolchain also allows for the deployment and configuration of the components that are required to execute these behaviors and automate the analysis of behaviors to ensure that they are safe and secure.
The development of CONVINCE is guided by three real-world scenarios, each posing unique challenges for robotic deliberations, which will be used to validate the results of the project.
• Robot Vacuum Cleaner (UC1): a robot that navigates in diverse home environments and should learn from its experience and avoid getting stuck.
• Assembly Robot (UC2): a robot that performs a complex task of assembling parts based on their shape, and should cope with occlusions, difficult lighting conditions, and missing or restrained parts.
• Robotic Museum Guide (UC3): a humanoid robot that guides visitors inside a museum and should interact with human and handle problematic situations that may arise from unpredictable human behavior.
In CONVINCE, a user-centered design approach is followed, where user needs are assessed to establish a set of high-level requirements. These requirements are then scrutinized concerning legal and ethical regulations, such as GDPR, Ethical AI, and the EU Artificial Intelligence Act, and connected to technical KPIs of Use Cases. Additionally, input from developers of robotic software, who are the end-users of the toolchain, is also taken into account and considered in CONVINCE to increase the success of the project and ensure the outputs meet the needs of the users.
The project has made significant progress towards the achievement of the goals set out in the proposal. Major results have been the first release of the toolchain and the completion of the first development iteration, which resulted in the validation of the use cases. Prerequisite for the implementation of the toolchain was a complete definition of the role of different components, the data they exchange and the formats for modelling tasks, components and skills, as well as the complete system under verification. Notably, the release of the toolchain includes working implementations of verification engines that have been tested on complete systems that represent simplified versions of the use cases.
CONVINCE will advance the state-of-the-art in three robotic software areas:
1) situation understanding and scene perception,
2) task and motion planning and
3) formal techniques for design-time and run-time verification.
CONVINCE adopts an open science strategy to maximize its impact on the robotic community. This involves making scientific results and datasets available on Zenodo (https://zenodo.org/communities/convince-project(s’ouvre dans une nouvelle fenêtre)) whenever possible. The toolchain supports the ROS middleware to encourage adoption by the ROS community. The first release of the main components of the toolchain has been made available under an open-source license on Github (https://github.com/convince-project(s’ouvre dans une nouvelle fenêtre)).
Notably, the project has established a ROS Deliberation Community Group to coordinate the development of deliberation frameworks and libraries, as well as, scripting languages, editors and runtimes, and BT tooling within CONVINCE and across the broader community. The working group is currently exploring ways to standardize the semantics of BT control nodes, which will aid the CONVINCE toolchain in model-check the deliberation layer based on BTs. This is expected to result in the creation of common architectural concepts and framework mechanisms, facilitating the integration of different methods.
1) situation understanding and scene perception,
2) task and motion planning and
3) formal techniques for design-time and run-time verification.
CONVINCE adopts an open science strategy to maximize its impact on the robotic community. This involves making scientific results and datasets available on Zenodo (https://zenodo.org/communities/convince-project(s’ouvre dans une nouvelle fenêtre)) whenever possible. The toolchain supports the ROS middleware to encourage adoption by the ROS community. The first release of the main components of the toolchain has been made available under an open-source license on Github (https://github.com/convince-project(s’ouvre dans une nouvelle fenêtre)).
Notably, the project has established a ROS Deliberation Community Group to coordinate the development of deliberation frameworks and libraries, as well as, scripting languages, editors and runtimes, and BT tooling within CONVINCE and across the broader community. The working group is currently exploring ways to standardize the semantics of BT control nodes, which will aid the CONVINCE toolchain in model-check the deliberation layer based on BTs. This is expected to result in the creation of common architectural concepts and framework mechanisms, facilitating the integration of different methods.