Periodic Reporting for period 2 - REGO (Cognitive robotic tools for human-centered small-scale multi-robot operations)
Período documentado: 2023-10-01 hasta 2025-03-31
Nature teaches us that the coordinated motion of multiple simple agents can enable complex tasks otherwise impossible to achieve. A remarkable example is given by ants. Indeed, a colony of garden ants can transform a pile of dirt into a multi-level underground structure in about a week, without a blueprint or a leader. Scientists have often taken inspiration from the coordination capabilities of animals to develop multi-robot systems showing bio-inspired collaboration and interaction features. The most striking results in this respect have been achieved by multi-drone systems. Nonetheless, such remarkable and often spectacular results achieved at the macro-scale did not translate well to smaller scales, where robotic systems, despite the promising features and highly-anticipated impacts, still face significant challenges in providing real added value.
Such daunting situation of small-scale robotics is due to different factors, including:
* Lack of interaction intelligence;
* Low cognitive capabilities;
* Low adaptability, scalability, and modularity;
* Limited capabilities and functionalities;
* Insufficient trust by the public.
We believe in a scientific revolution brought by AI-powered small-scale wireless multi-robot systems, naturally controlled by humans through advanced cognitive human-robot interaction techniques. We want to bring what is possible nowadays with multi-robot macro-scale robots to the smaller scales, powering small-scale robots with a deeper kind of cognitive intelligence, endowing them with higher dexterity and interaction capabilities with other robots, humans, and the environment.
Such daunting situation of small-scale robotics is due to different factors, including:
* Lack of interaction intelligence;
* Low cognitive capabilities;
* Low adaptability, scalability, and modularity;
* Limited capabilities and functionalities;
* Insufficient trust by the public.
We believe in a scientific revolution brought by AI-powered small-scale wireless multi-robot systems, naturally controlled by humans through advanced cognitive human-robot interaction techniques. We want to bring what is possible nowadays with multi-robot macro-scale robots to the smaller scales, powering small-scale robots with a deeper kind of cognitive intelligence, endowing them with higher dexterity and interaction capabilities with other robots, humans, and the environment.
The REGO project has evolved in line with the objectives and expectations. From the scientific point of view, the project has seen the continuation of WP1 on the “Design of small-scale stimuli-responsive wireless robots”, WP2 on the “Low-level wireless multi-robot control,” of WP3 on “High-level cognitive-based control and human-machine interfacing.” And the start of the WP4 on the applications. In this respect, the scientific work has focused on the design and development of magnetically-driven carrier robots, magnetic-field-driven small-size robots, as well as on their modelling and characterization. In parallel, we worked on the control capabilities of the distributed magnetic actuation system for advanced steering of the composite robotic systems at small scales, as well as designing its proper interfacing towards the other work packages. Finally, we have started some work with respect to the development of the multi-functional haptic handle.
We delivered two magneto‐mechanical carrier designs, one using two internal permanent magnets (IPMs) and one using three, each integrating modules for locomotion, payload storage, release, and local control. The two-magnet variant (15.5 × 6.5 mm and 14.6 × 4.2 mm prototypes) holds 47 µL (large) and 34 µL (small) of cargo, while the three-magnet variant (18 × 4.2 mm and 15 × 6.5 mm) holds up to 83 µL. Release distances from the triggering permanent magnet (r-EPM) range from 7 to 14 mm for the two-magnet design and 10 to 15 mm for the three-magnet design. Release efficiency for magnetic cargo measured 59 ± 9.3 % (two-magnet) versus 45.9 ± 9.2 % (three-magnet), with non-magnetic cargo coefficients of 48.3 ± 10.7 % and 55.3 ± 8.8 %, respectively. Both designs float under magnetic fields up to 99–140 mm away, demonstrating remote navigation. Finite-element and analytical models predict reliable actuation and guided release, and guided‐locomotion tests in water and glycerol validate rolling speeds from 4 to 136 mm s⁻¹ depending on medium and drive frequency. Model‐validated miniaturization yields functional carriers as small as 10.6 × 1.2 mm (two-magnet) and 22.9 × 3.2 mm (three-magnet), establishing clear design envelopes for application‐specific deployments., the participation to large public events reaching local communities and engaging in clustering activities with other projects.
A novel mobile hybrid magnetic actuation system, combining a robot-mounted electromagnet with rotatable permanent magnets, can generate up to 48 mT fields over human-scale workspaces, enabling precise steering of microrobots with minimal energy consumption. To overcome the rapid decay of magnetic gradients with distance, a tunable magnetic trap integrates Helmholtz coils and ferrite-rod passive elements, producing controllable local gradients that levitate and displace particles at fields below 5 mT. Locally actuated compliant grippers embed miniature planar coils around magnets to achieve high-force grasping of variable-stiffness polymer structures, supporting endoscopic and catheter-based manipulation. On the algorithmic side, a hybrid reinforcement-learning–Dynamic Window Approach framework enables real-time 3D navigation of nine-degree-of-freedom robot swarms in constrained environments, dynamically tuning local path planners for robust obstacle avoidance and efficient coverage. Auxiliary methods include a graph-neural-network tracker for precise 3D localization of multiple agents from sparse image inputs , and autonomous collaborative grasping and sorting strategies that use dipole-dipole and gradient forces to pick, place, and classify micro-objects with sub-0.3 mm precision.
The development of an ergonomic, multimodal haptic interface has started. Initial trials with an ultrathin polymeric‐coil interface revealed insufficient amplitude, prompting integration of four coin-shaped eccentric-rotor vibratory motors and Peltier cells into a Haption Desktop handle. Iterative mechanical refinements and user studies (n = 10) determined a just-noticeable displacement threshold of ~4 mm and informed the actuator layout . Vibratory feedback is driven open-loop, while thermal cues use PI-regulated, water-cooled Peltier control models to achieve <3% overshoot and ~3.1 s rise time. In parallel, we worked on designing a cognitive shared-control paradigm that blends user inputs with autonomous commands: four schemes allocate subsets of degrees of freedom between operator and AI to optimize task speed, collision avoidance, and space utilization in pipe-like environments, reducing subjective workload by up to 40% (NASA-TLX) compared to full manual teleoperation. Finally, complementary survey papers on cutaneous haptics and wearable multisensory devices have shaped design and control requirements.
Project advancements and research has been communicated and disseminated through different channels, including the project website and social media channels (250+ followers), by the participation in scientific and industrial conferences (15+ events), the publication of scientific papers (17 works already published, more under review – all in peer-reviewed international venues), the participation to large public events reaching local communities (3000+ people reached) and engaging in clustering activities with other projects (notably, HARIA). Finally, we have prepared a set of promotional materials for the partners to use when communicating about REGO.
A novel mobile hybrid magnetic actuation system, combining a robot-mounted electromagnet with rotatable permanent magnets, can generate up to 48 mT fields over human-scale workspaces, enabling precise steering of microrobots with minimal energy consumption. To overcome the rapid decay of magnetic gradients with distance, a tunable magnetic trap integrates Helmholtz coils and ferrite-rod passive elements, producing controllable local gradients that levitate and displace particles at fields below 5 mT. Locally actuated compliant grippers embed miniature planar coils around magnets to achieve high-force grasping of variable-stiffness polymer structures, supporting endoscopic and catheter-based manipulation. On the algorithmic side, a hybrid reinforcement-learning–Dynamic Window Approach framework enables real-time 3D navigation of nine-degree-of-freedom robot swarms in constrained environments, dynamically tuning local path planners for robust obstacle avoidance and efficient coverage. Auxiliary methods include a graph-neural-network tracker for precise 3D localization of multiple agents from sparse image inputs , and autonomous collaborative grasping and sorting strategies that use dipole-dipole and gradient forces to pick, place, and classify micro-objects with sub-0.3 mm precision.
The development of an ergonomic, multimodal haptic interface has started. Initial trials with an ultrathin polymeric‐coil interface revealed insufficient amplitude, prompting integration of four coin-shaped eccentric-rotor vibratory motors and Peltier cells into a Haption Desktop handle. Iterative mechanical refinements and user studies (n = 10) determined a just-noticeable displacement threshold of ~4 mm and informed the actuator layout . Vibratory feedback is driven open-loop, while thermal cues use PI-regulated, water-cooled Peltier control models to achieve <3% overshoot and ~3.1 s rise time. In parallel, we worked on designing a cognitive shared-control paradigm that blends user inputs with autonomous commands: four schemes allocate subsets of degrees of freedom between operator and AI to optimize task speed, collision avoidance, and space utilization in pipe-like environments, reducing subjective workload by up to 40% (NASA-TLX) compared to full manual teleoperation. Finally, complementary survey papers on cutaneous haptics and wearable multisensory devices have shaped design and control requirements.
Project advancements and research has been communicated and disseminated through different channels, including the project website and social media channels (250+ followers), by the participation in scientific and industrial conferences (15+ events), the publication of scientific papers (17 works already published, more under review – all in peer-reviewed international venues), the participation to large public events reaching local communities (3000+ people reached) and engaging in clustering activities with other projects (notably, HARIA). Finally, we have prepared a set of promotional materials for the partners to use when communicating about REGO.