ETH Zurich's breakthrough: Robots master opening doors with minimal human assistance

Numerous videos have showcased robots proficiently opening doors and navigating through them. However, many of these demonstrations have required significant human assistance.

This assistance can be in the form of remote control or guided training, where a human operator guides the robot through the process either in real time or during training for later replication. However, new research from ETH Zurich introduces a novel approach that aims to reduce the need for extensive human involvement.

Edging closer to truly 'Autonomous'

This approach involves three primary steps. First, the user describes the scene and the intended action. Second, the system devises a complex route. Lastly, this route is refined into an efficient path. The research paper explains that by offering high-level descriptions of both the robot and the object involved, along with a sparse objective outlining the task, the planning system comprehensively determines various aspects.

These include the robot's movement, the forces it should exert, the limbs it should utilize, and the timings and locations for establishing or breaking contact with the object. The system is divided into two main categories: object-centric and robot-centric. Object-centric tasks encompass actions like opening doors or dishwashers, while robot-centric tasks involve moving the robot around objects.

Other robots to adopt design

The research team highlights that this system can be adapted for various robot designs. While the demos focus on a quadruped robot, particularly ANYbotics' ANYmal, the research spin-off from ETH Zurich, the methodology can be applied to different robot shapes.

The team believes that this work can act as a foundation for eventually creating a fully autonomous system capable of locomotion and manipulation. This advancement takes us closer to achieving systems that can autonomously open doors without human intervention.