PhD - Computer Vision for Legged Robots

PhD - Computer Vision for Legged Robots

Tutors: Geoff Fink and Claudio Semini

Department: Dynamic Legged Systems (DLS) IIT, Genova
https://dls.iit.it

Description:
DLS is currently working on the implementation of computer vision (CV) algorithms for a range of diverse applications for quadruped robots (stair climbing, terrain mapping, obstacle avoidance, visual servoing (VS), automatic sensor calibration, state estimation, among others). Autonomous legged robots are required to handle a wide range of tasks in complex environments.

Current CV algorithms are not robust in dynamic environments, however, using CV is critical to improving autonomy. It is well known that video images provide rich information about the environment which is critical for localization in environments without a priori maps. For example, such uncertainty in environment would be expected in monitoring a disaster scene.

We are currently looking for a highly motivated, talented PhD student to work on the design, implementation, evaluation, and further development of CV [1,2] and state estimation algorithms~[3] into our legged robot framework. Tasks include:

  • Analyze the observability of legged robots with different sensor modalities
  • Explore ways to improve autonomy using VS, obstacle avoidance and SLAM
  • Explore ways to improve state estimation using CV algorithms and sensor fusion
  • Explore ways to improve the accuracy and robustness of current algorithms
  • Design algorithms to increase computation efficiency, code readability, and reusability
  • Experimental verification

Requirements:
The candidate should have a background in computer science, systems and control, robotics, or related. Further requirements include: strong programming skills in C++ and Python, knowledge of CV including multiple view geometry, strong communication skills (written and spoken) in the English language, strong team player, knowledge about robot kinematics and dynamics, VS, SLAM, and control theory, hands-on experience in robotic systems is a big plus, programming skills in OpenCV, CUDA, and Matlab/Simulink is a plus.

*Note: It is compulsory to prepare a research proposal on this topic.

References:
1. M. A. Bloesch, State estimation for legged robots–kinematics, inertial sensing, and computer vision. PhD thesis, 2017.
2. F. Chaumette and S. Hutchinson, “Visual servo control. ii. advanced approaches [tutorial],” IEEE Robotics Automation Magazine, vol. 14, pp. 109–118, March 2007.
3. S. Nobili, M. Camurri, V. Barasuol, M. Focchi, D. G. Caldwell, C. Semini, and M. Fallon, “Heterogeneous sensor fusion for accurate state estimation of dynamic legged robots,” in Robotics: Science and Systems XIII, MIT, Cambridge, Massachusetts, July, 2017.

Contacts:
Email: geoff.fink@iit.it and claudio.semini@iit.it