Research
Path replanning
The research aims to develop a path replanning algorithm for robots in dynamic environments. In human-robot collaboration, the robot must react promptly to the operator's movements to avoid collisions. Slowing the robot when the operator approaches is common but inefficient. Instead, changing the robot's path on the fly allows it to avoid moving obstacles and reach the goal safely. The main challenge is to compute new trajectories quickly that the robot can continue toward the goal without stopping at every obstruction. In addition, the research aims to integrate a human-aware cost function so that the replanning algorithm finds paths that maintain a minimum distance from the operator, increasing his confidence in the robot.
Contact: Cesare Tonola
Game-theory for Human-Robot interaction
The research is focused on physical Human-Robot Interaction. In particular, Game-theoretical models (cooperative and non-cooperative) are used to describe the interaction and define robot control actions accordingly. Different models allow different robot behaviors (assistive, repulsive), and switching between them provides for Role Arbitration.
Finally, since knowledge of human behavior and modeling is necessary for problem formulation, part of his research focuses on human control modeling and human motion intention prediction.
Contact: Paolo Franceschi
Reinforcement Learning for industrial robot programming
The research focuses on making industrial robot programming a tool for any user. Currently, industrial programming methods require robotics and programming knowledge from the user, and only specialized operators can program a robotic task. Some intuitive programming methods have been developed but present critical issues for the industrial world. They require dedicated environments or tools. This research aims to create a framework that allows the user to program using a high-level language that hides complex operations requiring high knowledge.
Reinforcement learning with a simulated environment is the tool that we want to use to generate a complex task program starting from the base information that an ordinary user can provide.
Contact: Michele Delledonne
Multi-Agent Coordination and Planning
The research addresses the challenge of multi-agent (human-robot and multi-robot) coordination in collaborative work cells.
TAMP (Task and Motion Planning) methodology is a classic approach to integrate task and motion planning problems in this research field.
In order to ensure efficient and safe coordination at the task planning level, the aim is to formalise "Human-Aware Task and Motion Planning" techniques. Different methodologies are being tested:
Task Planning as a "Mixed Integer Linear Programming" optimisation problem by considering the quality of the interactions between agents (online estimation).
Task Planning as a Sequential decision-making problem solved with Reinforcement Learning.
Contact: Samuele Sandrini
Robothon® - The Grand Challenge : A Global Robotics Competition in Advanced Manipulation
Robothon® is an international competition and benchmarking event for measuring state-of-the-art performance in robot manipulation.
CARI young researchers participated and won the 2022 Robothon Competition.
The challenge is about the disassembly and sorting of electronic waste and the goal is to make these processes even more efficient and sustainable through automated recycling. Accomplishing this requires enabling robots to autonomously recognize, grasp, and manipulate loose components in a new production environment.
Visit here the competition site to read more informations.
Award Ceremony
The award ceremony took place at the "automatica" fair (Trade Fair Center Messe München) on the munich_i platform.
The CARI team consisting of Cesare Tonola, Samuele Sandrini, Michele Delledonne and Roberto Pagani received the first prize financed by Huawei and presented by Prof. Dr.-Ing Juergen Grotepass (Chief Strategy Officer Manufacturing Huawei Technologies).
Proposed Solution
A short video shows the methodologies adopted for solving the tasks proposed by the competition.
The final part of the tape is a demo of transferability from the task board to the real world: we chose to apply them to the disassembly of dismissed PCs.