Short Bio

I am a third year PhD student working on humanoid robots at DIAG Robotics Laboratory with Professor Giuseppe Oriolo. I am mainly interested in locomotion on uneven terrain, motion planning and control algorithms. I am currently studying efficient techniques to make humanoids climb stairs quickly and safely.

Projects

For the last two years I have been working mainly on humanoid robot locomotion, focusing in particular on motion planning, state estimation, mapping and control. Here is a list of the main projects I have developed.

Planning and Executing Humanoid Gaits
in a World of Stairs

During my master thesis I developed a pipeline for humanoid robot locomotion in unknown environments using terrain mapping, footstep planning and variable height MPC. The project has been developed in C++ and Python using ROS and the B-Human framework. In the end I performed multiple experiments with a NAO humanoid robot.

Research

Currently, my focus is on efficient motion planning with applications to humanoid robot navigation.

Publications

List of publications not yet available.

Teaching

I am currently a supervisor for the final projects of Autonomous and Mobile Robotics.
Here is the list of the projects I have assigned during the past years.

Autonomous and
Mobile Robotics

Autonomous humanoid navigation in large-scale environments

Monica De Pucchio, Elisa Foderaro, Francesco Petri

Autonomous navigation in large-scale environments remains a major challenge in robotic systems. Recent research ideas presented during the DARPA SubT Challenge showed that it is possible to explore and navigate complex unknown environments using both ground and aerial vehicles. The aim of this project is to extend our humanoid navigation framework (Ferrari et al.) by implementing a global-planner (Dang et al.) and by integrating it with our RRT*-based footstep planner, which will act as a local planner. Simulations must be performed to validate the implementation. This project must be implemented in C++.

References
  • Dang et al., “Graph-based Subterranean Exploration Path Planning using Aerial and Legged Robots”, Journal of Field Robotics 2020
  • Ferrari et al., “An Integrated Motion Planning/Controller for Humanoid Robots on Uneven Ground”, ECC 2019

Planning parking maneuvers for a car-trailer vehicle

Leandro Maglianella, Lorenzo Nicoletti, Olga Sorokoletova

Being a nonholonomic vehicle, planning parking maneuvers for a car-trailer vehicle requires accounting for the kinematic constraints on the system. In sampling-based planners like RRT, this can be done by sampling the input space and forward integrating the system generating dynamically feasible motion primitives. The car-trailer system is however unstable in backward motion, exhibiting the so-called “jackknifing” phenomenon, in which the hitch angle between the car and the trailer heading angles diverges, leading to a loss of maneuverability and ultimately to self collision. For this reason, planning backward trajectories requires special care. The project consists in the implementation of an RRT-based planner for the car-trailer vehicle using the Open Motion Planning Library (OMPL), possibly implementing the technique presented in the accompanying paper to stabilize backward motion primitives. The project must be implemented in C++. This project has been supervised together with Tommaso Belvedere.

References
  • Evestedt et al, “Motion planning for a reversing general 2-trailer configuration using Closed-Loop RRT”, IROS 2016

Improving footstep planning algorithms by efficient nearest neighbor searching

Ionut Marian Motoi, Leonardo Saraceni, Giuseppe Sensolini Arrà

One of the bottlenecks in the performance of sampling-based motion planning algorithms is the computational cost of the nearest-neighbor operation. It is, hence, essential to develop efficient techniques for nearest-neighbor searching in order to keep the planning time low. The aim of this project consists in extending the RRT and RRT* footstep planners proposed by our group by implementing a self balancing k-d tree on non-Euclidean topologies. Simulations on V-REP/CoppeliaSim must be performed to validate the method. This project must be implemented in C++ using ROS and our footstep planning framework.

References
  • Yershova et al, “Improving Motion Planning Algorithms by Efficient Nearest-Neighbor Searching”, T-RO 2007
  • Ferrari et al, “An Integrated Motion Planning/Controller for Humanoid Robots on Uneven Ground”, ECC 2019