TL;DR
This paper presents a novel, fully scalable, multi-modal sensor system for robotic and prosthetic hand fingers. By combining a parametric CAD model with modular, commercially available sensors, they've created a versatile system capable of sensing forces, distance, acceleration, and joint angles, paving the way for more dexterous and human-like artificial hands.
What Problem Does This Paper Address?
Grasping and manipulating objects is a fundamental challenge for both robotic and prosthetic hands. While visual feedback is crucial, a sophisticated sense of touch is essential for fine manipulation and robust interaction with the environment. However, embedding the necessary sensors into the complex and confined mechanical structure of artificial fingers, while maintaining scalability for different hand sizes, has been a significant hurdle. This paper tackles the problem of integrating a rich set of sensing modalities into a mechanically scalable and electronically modular finger design.
Methods
The core of the paper lies in its parametric, skeleton-based CAD model for the finger. This model allows for the automatic scaling of finger dimensions to match human hand sizes, a critical feature for prosthetic applications. The mechanical parts are produced using 3D printing.
The multi-modal sensor system is built using interchangeable modules of commercially available sensors. This includes:
- Normal and Shear Force Sensors: Two types are used: barometer-based sensors (sensitive to low forces but saturate quickly) and Hall effect-based sensors (wider range but more non-linear behavior).
- Distance Sensor: A time-of-flight (ToF) sensor for pre-grasp control and object detection.
- Accelerometer: For slip detection.
- Joint Angle Encoders: Using off-axis Hall effect sensors and magnets to measure joint rotation.
- Temperature Sensors: Integrated into most sensor chips.
The electronic components are designed for modularity and minimal wiring, using flat flex cables (FFCs) and I²C communication to reduce complexity and space requirements. The manufacturing process involves gluing PCBs, casting sensors in silicone rubber, and then encapsulating the entire fingertip in a final silicone layer for robustness and compliance.
Datasets
Not discussed in the paper. The paper focuses on the design and experimental evaluation of the sensor system itself, using controlled laboratory experiments rather than large datasets for training.
Key Insights and Ideas
- Scalability is Key: The parametric CAD model is a significant contribution, enabling the creation of fingers that can be tailored to individual human hand dimensions, a crucial aspect for prosthetic applications.
- Modularity for Versatility: The use of interchangeable sensor modules allows for easy adaptation of the sensor suite to different applications and finger sizes, making the system highly flexible.
- Sensor Fusion for Robustness: The paper demonstrates that detecting events like slip is more reliable when data from multiple sensor modalities (force, acceleration, joint angle) are fused, providing a more confident assessment.
- Hysteresis in Silicone: The experiments revealed that the silicone encapsulation, while providing robustness and compliance, introduces hysteresis in the sensor readings, particularly for force sensors. This is an important consideration for precise control.
- Trade-offs in Sensor Choice: The paper highlights the trade-offs between different sensor types, such as the sensitivity vs. range of barometer-based vs. Hall effect-based force sensors.
Key Results
- The developed parametric model was successfully tested for both 5th percentile female and 95th percentile male hand dimensions.
- The sensor system demonstrated the ability to track normal and shear forces, detect distance, and measure joint angles.
- Experiments showed that distinct events during grasping (contact, slip) could be detected by multiple modalities, enabling more confident event detection through sensor fusion.
- The accelerometer proved valuable for gross slip detection, even when damped by soft material.
- The spatial resolution of the fingertip sensors was found to be sufficient for grasping and lifting tasks, though higher sensor density might be needed for fine manipulation.
- The manufacturing process, utilizing 3D printing and commercially available sensors, allows for individual finger production at a cost below 70€.
Limitations and Future Work
- Hysteresis: The silicone encapsulation introduces hysteresis, which needs to be accounted for in control strategies.
- Sensor Placement: While the current sensor density is deemed sufficient for many tasks, higher density might be required for fine-grained manipulation.
- Crosstalk: While minimal crosstalk was observed between sensors, it's an area that could be further investigated.
Future work includes:
- Integrating the fingers into ongoing work on hand prostheses and humanoid robotic hands to further test robustness and validate the design.
- Exploring different sensor fusion approaches to extract semantic information from the high-dimensional sensor data.
- Investigating incipient slip detection more thoroughly, potentially by examining accelerometer mounting positions.
- Considering the addition of further electronics and sensing modalities, such as regulated heating elements for temperature flux measurements.
Summary updated