New tech enables robots to handle complex, irregular objects

Handling irregular and complex-shaped objects has long been one of the biggest challenges in robotics. Current systems, which rely on large databases and rigid models, struggle to adapt to unpredictable tasks and environments. However, a breakthrough introduced by researchers at the Swiss Federal Institute of Technology in Lausanne (EPFL) and the Idiap Research Institute opens new doors for industrial and service robots.

This new approach enables robots to manipulate, cut, and peel any object, even in cluttered settings with incomplete data. The core of the system is based on adaptable geometry, creating a “point cloud map” of any object regardless of its shape or size. This map identifies key surface points and generates a smooth geometric representation tailored to the task.

As Interesting Engineering reports, this structure helps robots understand and handle objects with varying geometries, overcoming limitations of traditional methods. What’s innovative is that it doesn’t require perfect models or large training datasets. Instead, it uses principles of discrete differential geometry to build local reference frames guided by the surface and key points of the object.

These orientation fields make it possible for robots to perform basic actions—like sliding, cutting, or peeling—invariant to the object’s shape. This allows skill transfer between very different objects, mimicking human flexibility. The technology addresses the variability in geometry that often hampers robotic manipulation.

The system can be integrated with various control schemes, from teleoperation to trajectory planning and reinforcement learning. By generating real-time orientation fields from vision and depth data, robots can quickly adapt to new objects and situations, boosting their efficiency and versatility.

The key to this breakthrough is generating an orientation field from real-time vision and depth data.