To get updates from our aUToronto self-driving car team, please visit this page.
Dynamic Systems Lab News
Congratulations to Melissa Greeff for successfully defending her Ph.D. in “Flying Flat Out: Fast Multirotor Flight in Real-World Environments”! Many thanks to the external reviewer Prof. Vijay Kumar.
Melissa made strong theoretical contributions around differential flatness and learning-based control and applied these concepts to the challenging robotic task of outdoor autonomous vision-based multirotor flight. Her work was supported by NSERC, DRDC and Drone Delivery Canada.
She will be joining as a faculty member at Queen’s University in September. Check out her work and talk below:
Departmental Seminar: www.youtube.com/watch?v=ubRcyOShHIg
Google Scholar: tiny.cc/greeff_gs
Webpage: roboralab.com/index.html ... See MoreSee Less
SiQi Zhou successfully defended her PhD in “Neural Networks as Add-on Modules for Improved Performance of Robot Control Systems.” Thanks to all committee members, Prof. Tim Barfoot, Prof. Jonathan Kelly, Prof. Aude Billard, and Prof. Florian Shkurti, for their valuable feedback!
In her thesis, SiQi proposed novel learning-based control approaches that safely and efficiently combine control theory and machine learning techniques for designing high-performance robot control systems. In particular, she demonstrated how we could leverage the modeling capability of neural networks as add-on blocks to enhance the performance of classical control systems while leveraging control theory to guide safe and efficient deployment. The works presented in her thesis were demonstrated in quadrotor impromptu tracking experiments and flying inverted pendulum experiments. Check out her DDS talk here for details tiny.cc/zhou-phd-video
Google Scholar: scholar.google.ca/citations?user=5fGdja8AAAAJ&hl=en&oi=ao
University of Toronto Engineering ... See MoreSee Less
When deploying robots to the real world, providing safety guarantees in the presence of uncertainties is challenging. Our #L4DC2022 work presents a Barrier Bayesian Linear Regression (BBLR) technique for online learning safety conditions for uncertain systems.
In this work, we propose a BBLR model exploiting the structure of the control barrier condition and learn this condition in the safety filter from online data of an uncertain robot system. The proposed BBLR safety filter allows us to guarantee safety during the learning process.
Our paper can be found at this link: proceedings.mlr.press/v168/brunke22a/brunke22a.pdf
We will present our work at #L4DC this Thursday (June 23) @ 16:15-17:15 PT, poster number 12. We look forward to seeing you there!
Angela Schoellig ... See MoreSee Less
Thanks to the Bitcraze team for featuring our work in their latest blog post, highlighting our work about robust UWB localization and our recent UTIL (Ultra-wideband Time-difference-of-arrival Indoor Localization) dataset.
Blog Post: tinyurl.com/436mhep3
Ultra-wideband (UWB) time-difference-of-arrival (TDOA)-based localization has recently emerged as a promising, low-cost, and scalable indoor localization solution for multi-robot applications. However, there appears to be a lack of public datasets to benchmark the emerging UWB TDOA positioning technology in cluttered indoor environments. To fill this gap, we released the UTIL (Ultra-wideband Time-difference-of-arrival Indoor Localization) dataset based on the Loco Positioning System from Bitcraze. Our dataset consists of (i) an UWB identification dataset in a variety of occluded scenarios involving obstacles with different types of materials and (ii) a comprehensive multi-modal dataset collected with a cumulative ~150 minutes of real-world flights in indoor environments cluttered with static and dynamic obstacles. We hope this dataset can foster the research in UWB TDOA-based positioning in realistic challenging environments.
We would like to thank Kristoffer Richardsson and Tobias Antonsson from Bitcraze for their guidance on the software and hardware development during the dataset creation.
Angela Schoellig ... See MoreSee Less
Maintaining an up-to-date map to reflect changes in the scene is very important, as failing to detect and correct such changes can cause a deterioration in map quality, leading to poor localization and lost robots. However, these incremental changes are not always easy to detect with existing mapping systems.
We have been working on a solution to this problem! Our work on mapping under semi-static scene changes has been accepted to #RSS2022! We show that our framework produces more consistent maps despite the scene changing significantly over time, and is robust to sensor noise, partial observations and dynamic objects. We summarize our work with the following contributions:
- We introduce a Bayesian update rule to propagate each object's stationarity and pose-change assessment using both geometric and semantic information
- We present a novel online, object-aware map maintenance framework leveraging the proposed Bayesian object-level update rule
- We release a real-world change detection dataset captured using a robot platform in a warehouse setting
Thank you to Clearpath, Nserc Canada and the Vector Institute for supporting our research. We look forward to seeing everyone at #RSS2022 soon!
Check out our paper and dataset for more details.
arXiv preprint: arxiv.org/abs/2205.01202
Dataset: github.com/Viky397/TorWICDataset ... See MoreSee Less
Join us on Friday in room 113C for our #ICRA2022 workshop “Releasing Robots into the Wild: Simulations, Benchmarks, and Deployment” to see some amazing talks and discussions!
Angela Schoellig University of Toronto Engineering ... See MoreSee Less
We are finally able to be at ICRA in-person. We hope to see you!
Prof Schoellig’s Talk @ Tutorial: Tools for Robotic Reinforcement Learning (Monday 13:30-14:30) lnkd.in/gm_Y2sqm
Fly out the window: exploiting discrete-time flatness for fast vision-based multi-rotor flight @ Aerial Robotics I (TuA01.05, 10:20-10:25)
Bridging the model-reality gap with Lipschitz network adaptation @ Learning for Control (ThB10.04, 15:45-16:50)
Workshop on Releasing Robots in the Wild (Friday 08:30-17:00)
lnkd.in/ehVZb2r2 ... See MoreSee Less