Research

My research interest is mainly focused on visual perception for robotic application. I use bio-inspired and conventional methods to solve robotic tasks such as stabilization, pursuit and obstacle avoidance.

Event-based visual sensor

A nice description of how the sensor works can be found here

• Work presenting an extension from the IBM DVS gesture dataset to perform gesture classification from event-based vision sensor. The addition of a detection module allows to limit the complexity of the classifier and to run on a neuromorphic hardware, while being robust to variation of distances and light conditions. It also make it possible to have multiple subject in the frame.
  

• Event-by-event pose estimation based on a 3D point matching and EKF state estimation. The algorithm was also adapted for multiple sensors and with mapping it shows good accuracy on the tested environment.
  

• The Davis240C sensor was used to compute Optic Flow and perform obstacle avoidance with a quadrotor aat up to 2.5m/s.
  

Insect inspired vision and application

The following research was made during my PhD in the Biorobotics team in Marseille. My thesis can be downloaded here: Link to thesis .

• Using an Artificial compound eye submitted to periodic micro-movements displays hyperacuity, i.e. locating features with a greater accuracy than that corresponding to the resolution imposed by the photoreceptor’s limited pitch. Application of this principal over a large field of view allowed a robot to stabilized is position or follow a target.
  

  • Robotic visual stabilization and short-range odometry

  • Robotic visual pursuit under different light conditions

    
    which mimics the behavior of hoverfly trajectories




• Collaboration on the robot LORA, which aims at replicating bees navigation into a large corridor. The different behaviors such as centering, wall-following and speed regulation in a tappered corridor were reproduced using Optic Flow measurements.