Heron@CNR Joint Lab

Vision

Despite its remarkable success Robotics and AI may face serious bottlenecks in a not so far future. The prevailing mechatronic design paradigmis affected by serious issues, , even when it is integrated and augmented by Machine Learning and Probabilistic Robotics technologies. Those doubts have surfaced also on general public media, for example on the New Scientist.

Within the mechatronic paradigm approach if you want to implement more complex behaviors you will have the complexity of your system growing more quickly. We need to go beyond mechatronics, as it was aimed by the past RoboCom Flagship proposal and the new RoboCom++ FET-Flagship Proof-of-concept project, or as it has been preached for several years by the ShanghAI Lectures.

In the current approach control is top down and the ‘body’ and the ‘mind’ of the robot are well divided, following the Cartesian vision. In nature intelligent agents follow a completely different paradigm based on the self-organization of loosely connected networks of embodied agents. The next generation of robots will mimic the principle of organization of natural intelligent agents. We aim to make this vision real.

We aim to make this vision real.

Challenges

The control of robot manipulators in the real world is highly challenging due to uncertainty related to the estimation of the system state, the non-linear underlying dynamic and the imperfect perception of the surrounding environment that can jeopardize the manipulator efﬁciency leading the robot to a failure in the assigned task. Underwater environment is a particularly critical scenario in which quickly changing conditions and noisy observations affect the state estimation of the manipulator and cause difﬁculties in ﬁnding and grasping objects.
Uncertainty of robotic systems generally derives from three main sources: (i) motion uncertainty (also known as process uncertainty) resulting from noise affecting system dynamics, (ii) sensing uncertainty (also referred to as non-perfect state information) caused by noise jeopardizing sensor measurements, and (iii) environmental uncertainty, consisting in uncertain location of obstacles or features. Motion planning strategies should be effective and robust enough to overcome all these issues.
Non-conventional algorithms are cutting-edge solutions to be explored in order to make mobile robots effectively deal with real world situations, given the high degree of uncertainty and unpredictability. In real applications, in which sensor measurements are affected by significant errors, such non-conventional techniques allow to plan the desired reaching movement with the two concurrent objectives of reducing: i) the end-effector distance from the target and ii) the uncertainty on the measure.
The approach can be easily extended to lighter soft robotics arms, characterized by uncertain kinematics and dynamics. Such techniques are promising in order to obtain effective behavior and significant autonomy for near- future robots, and they will allow to mark a substantial step-change in the domain of field robotics: I envision a future world in which robots safely work side by side with humans, both in domestic and in civilian and industrial scenarios, relieving men and women of hard, dangerous and repetitive tasks.
Soft mechanics: Conventional robots are characterized by rigid links, joints and end-effectors, i.e. manipulators and other tools for interacting with objects. Compliance with the environment is guaranteed through high performance tactile sensors and control systems. The introduction of mechanically soft links, joints, actuators and end-effectors will decrease the complexity and cost and increase the safety of the system (avoiding breaking of the handled objects, structure of the robot itself or actuators).

A scientific cooperation

Heron Robots

CNR-INM

Scope

Vision

People

Fabio Bonsignorio

Enrica Zereik

Challenges

Contacts

A scientific cooperation

Heron Robots

CNR-INM

Credits