Research
I am interested in how closing the loop between perception and action can improve performance in manipulation tasks for robotics.
My research focus is robot perception, machine learning and control.
|
>
|
Uncertainty averse pushing with model predictive path integral control
Ermano Arruda*, Michael J Mathew*, Marek Kopicki, Michael Mistry, Morteza Azad, Jeremy Wyatt (* equal contribution)
International Conference on Humanoid Robots (Humanoids), 2017
video
/
bibtex
This work introduces a planner that makes use of an uncertain, learnt forward (dynamical) model to plan robust push manipulation. The forward model of the system is learned by poking the object in random directions and is then utilised by a model predictive path integral controller to push the box to a required goal location. By utilising path integral control, the proposed approach is able to avoid regions of high predictive uncertainty in the forward model. Thus, pushing tasks are completed in a robust fashion with respect to estimated uncertainty in the forward model and without the need of differentiable cost functions.
|
>
|
Active vision for dexterous grasping of novel objects
Ermano Arruda, Jeremy Wyatt, Marek Kopicki
International Conference on Intelligent Robots and Systems (IROS), 2016
video
/
bibtex
We tackled the problem of improving robot grasp performance using active vision. We sought to increase grasping success via two view selection heuristics: one that would allow the robot to explore good quality grasp contact points, and another that would permit the robot to investigate its workspace to make sure candidate grasp trajectories would not lead to collisions with unseen parts of the object to be grasped. Our results showed that this approach yielded better grasp success rate when compared to a random view selection strategy, while using fewer camera views for grasp planning.
|
|
- First place at ISMAR Off-site Tracking Competition, Fukuoka, Japan, 2015. Developed a monocular visual odometry system with additional sparse bundle adjustment for camera trajectory optimisation.
video1
/
video2
code
|
|
- Winning team CESAR-VoxarLabs at LARC/CBR - Latin American and Brazilian Robotics Competition, RoboCup@Home, 2014. Worked on object-tracking and detection system.
meet i-zak
|
|