Sergio
Picella

Soft robots, which are flexible and mimic living organisms, often require complex electronics and software to make decisions based on their environment. We introduce a dictionnary of pneumatic coding blocks instead of electronics to enable high-level, electronics-free coding of programs in fluidic soft robots.

These blocks mimic basic programming commands like If or For and rely on the natural mechanical properties of materials. By combining these blocks, we created a soft robotic gripper that can switch behaviors based on its surroundings.

This approach mirrors how living systems encode functions in their bodies, offering a more natural, efficient, and autonomous way to program soft robots for complex tasks. [video]

Soft robots can adapt to different environments by changing shape in response to forces, but sensory feedback is crucial for higher-level decision-making. Current sensing technologies require separate sensing elements integrated into soft actuators, complicating fabrication and reducing robustness.

We present a versatile sensing strategy that retrofits existing soft fluidic devices without design changes by measuring the fluidic input needed to activate a soft actuator and linking it to its deformed state. This approach enables tactile sensing of object size, shape, roughness, and stiffness, and demonstrates robustness in tasks like fruit sorting and ripeness detection, without requiring embedded sensors or redesign. [video]

Achieving adaptivity in complex environments is a significant challenge in robotics. Traditional systems rely on stiff materials and centralized controllers, while nature often employs soft materials and distributed intelligence. Our study introduces a decentralized approach for robust behavior in soft robotic systems that requires no prior knowledge of shape or environment.

By implementing simple rules in identical modules, the system achieves whole-body phototaxis—moving toward a light source — without explicit communication between modules. The results demonstrate robust, adaptive behavior in dynamic environments, highlighting the importance of system geometry and learning parameters. This approach offers a pathway to adaptive, autonomous robots requiring minimal computational power.