The DARKO project has made significant advancements across four main objectives: dynamic mobile manipulation, predictive safety and efficiency in human-robot coordination, efficient deployment and safe localization, and risk-aware operation for safety and efficiency. Here's a summary of the main work done and the key results in each area:
O1: Efficient and Safe Dynamic Mobile Manipulation
- Main Work: Enhancements in dynamic manipulation tasks were achieved through the development of inherently elastic manipulators, advanced high-speed perception capabilities, and algorithms to improve efficiency and safety.
- Key Results: Improved performance of the SoftHand, a new elastic wrist design, development of modular Bi-Stiffness Actuators (BSA), and comprehensive controller packages for the Franka Emika Panda. integration of BSA into an elastic robotic arm. Innovative methods (e.g.Thunder Dynamics library, adaptive torque control, learning-based and dictionnary-based methods) for robust throwing motions and human-like throwing strategies were also introduced. Finally, a 360-degree object detection with fish-eye camera and lidar input was developed for efficient detection of objects to be manipulated.
O2: Predictive Safety and Efficiency in Human-Robot Coordination
- Main Work: Focused on integrating robots into existing warehouse operations safely and efficiently, utilizing long-term human motion prediction methods, and developing a multi-modal motion capture dataset.
- Key Results: Enhanced human-robot interaction through advanced motion prediction algorithms, integration of gaze tracking technology, vocal interface, and the deployment of the Anthropomorphic Robot Mock Driver (ARMoD). Introduction of neuro-symbolic architecture for motion prediction and implementation of a human-aware navigation system. Development of the THOR-MAGNI Act dataset with action-level annotations for training activity aware models, even in occlusion-prone environments.
O3: Efficient Deployment and Safe Localization
- Main Work: Aimed at creating mapping and localization systems that are failure-aware and resilient, focusing on data-efficient perception methods to reduce the need for extensive semantic annotations.
- Key Results: Creation of an annotated 3D intralogistics dataset, deployment of the multi-class 9DoF RGB-D YOLO++ detector, and advancements in grasp pose estimation and object detection technologies. Significant improvements in human perception operations and the development of a novel neural surface representation for efficient and accurate 3D reconstruction. Development of a pipeline for real-time large-scale 3D mapping, efficient and consistent even with uncalibrated video footage.
O4: Risk-aware Operation for Safety and Efficiency
- Main Work: Incorporation of risk assessment into robotic decision-making, focusing on predicting and mitigating potential risks to ensure safety and operational efficiency.
- Key Results: Development of methodologies for anticipating navigation and manipulation risks, continuous monitoring (with causal reasoning algorithms) of risk levels during operations, and the implementation of a Stochastic Dynamic Programming engine (and task scheduler) to optimize decision-making under uncertainty. Development and integration of a Safe Motion Unit aligned with latest safety standards.
These achievements collectively enhance the capabilities of robotic systems in dynamic manipulation, human-robot interaction, autonomous navigation, and risk-aware operation, setting new standards for safety, efficiency, and technological innovation in the field of intralogistics robotics.
At the project end, two integrated mobile robotic platform have been developed and demonstrated to academic and industrial stakeholders, and have been showcased at a large open event (Automatica 2025, Munich). The system was capable of autonomously map the environment, interact efficiently with nearby humans, identify, grasp and throw objects to a dedicated target area. The consortium developed open-source libraries that can be used freely in the future. The project led to an italian patent, and extends the product portfolio of one of the industrial partners.
