- 79
- 63 985
Learning Systems and Robotics Lab
Canada
เข้าร่วมเมื่อ 3 ส.ค. 2018
The Learning Systems and Robotics Lab, headed by Prof. Angela Schoellig, envisions a new generation of intelligent robotic systems that seamlessly interacts with the physical world. In particular, our research interests are centered around the challenges associated with robots operating in increasingly unstructured, uncertain, and changing scenes. By conducting research at the intersection of robotics, controls, and machine learning, we enhance the performance, safety, and autonomy of robots navigating in human-centric environments.
Safety Filtering While Training: Improving the Performance and Sample Efficiency of RL Agents
Abstract:
Reinforcement learning (RL) controllers are flexible and performant but rarely guarantee safety. Safety filters impart hard safety guarantees to RL controllers while maintaining flexibility. However, safety filters can cause undesired behaviours due to the separation between the controller and the safety filter, often degrading performance and robustness. In this paper, we propose several modifications to incorporating the safety filter in training RL controllers rather than solely applying it during evaluation. The modifications allow the RL controller to learn to account for the safety filter, improving performance. Additionally, our modifications significantly improve sample efficiency and eliminate training-time constraint violations. We verified the proposed modifications in simulated and real experiments with a Crazyflie 2.0 drone. In experiments, we show that the proposed training approaches require significantly fewer environment interactions and improve performance by up to 20% compared to standard RL training.
Paper: tinyurl.com/sf-train-paper
Code: tinyurl.com/sf-train-code
Researchers:
Federico Pizarro Bejarano, Lukas Brunke, Angela P. Schoellig
For more:
www.dynsyslab.org/vision-news/
Reinforcement learning (RL) controllers are flexible and performant but rarely guarantee safety. Safety filters impart hard safety guarantees to RL controllers while maintaining flexibility. However, safety filters can cause undesired behaviours due to the separation between the controller and the safety filter, often degrading performance and robustness. In this paper, we propose several modifications to incorporating the safety filter in training RL controllers rather than solely applying it during evaluation. The modifications allow the RL controller to learn to account for the safety filter, improving performance. Additionally, our modifications significantly improve sample efficiency and eliminate training-time constraint violations. We verified the proposed modifications in simulated and real experiments with a Crazyflie 2.0 drone. In experiments, we show that the proposed training approaches require significantly fewer environment interactions and improve performance by up to 20% compared to standard RL training.
Paper: tinyurl.com/sf-train-paper
Code: tinyurl.com/sf-train-code
Researchers:
Federico Pizarro Bejarano, Lukas Brunke, Angela P. Schoellig
For more:
www.dynsyslab.org/vision-news/
มุมมอง: 92
วีดีโอ
Preventing Unconstrained CBF Safety Filters Caused by Invalid Relative Degree Assumptions
มุมมอง 101หลายเดือนก่อน
Title: Preventing Unconstrained CBF Safety Filters Caused by Invalid Relative Degree Assumptions Preprint: arxiv.org/abs/2409.11171 Code: github.com/lukasbrunke/multi-cbf Researchers: Lukas Brunke, Siqi Zhou, Angela P. Schoellig For More: www.learnsyslab.org/ Abstract: Control barrier function (CBF)-based safety filters are used to certify and modify potentially unsafe control inputs to a syste...
Optimized Invariance Conditions for Uncertain Input-Constrained Nonlinear Systems (Presentation)
มุมมอง 2382 หลายเดือนก่อน
Published in IEEE L-CSS 2023 and presented at ACC 2024 in Toronto. Title: Optimized Control Invariance Conditions for Uncertain Input-Constrained Nonlinear Control Systems Paper: ieeexplore.ieee.org/abstract/document/10365501 Preprint: arxiv.org/abs/2312.09565 Researchers: Lukas Brunke, Siqi Zhou, Mingxuan Che, Angela P. Schoellig For More: www.learnsyslab.org/ Abstract: Providing safety guaran...
ACC24: Is Data All that Matters? The Role of Control Frequency for Uncertain Systems (Presentation)
มุมมอง 1142 หลายเดือนก่อน
Presented at ACC 2024 in Toronto. Title: Is Data All that Matters? The Role of Control Frequency for Learning-Based Sampled-Data Control of Uncertain Systems Paper: ieeexplore.ieee.org/document/10644546 Preprint: arxiv.org/abs/2403.09504 Researchers: Ralf Römer, Lukas Brunke, Siqi Zhou, Angela P. Schoellig For More: www.learnsyslab.org/ Abstract: Learning models or control policies from data ha...
ACC24: Practical Considerations for Implementing CBF Safety Filters (Presentation)
มุมมอง 1742 หลายเดือนก่อน
Presented at ACC 2024 in Toronto. Title: Practical Considerations for Discrete-Time Implementations of Continuous-Time Control Barrier Function-Based Safety Filters Paper: ieeexplore.ieee.org/document/10644713 Preprint: arxiv.org/abs/2404.12329 Researchers: Lukas Brunke, Siqi Zhou, Mingxuan Che, Angela P. Schoellig For More: www.learnsyslab.org/ Abstract: Safety filters based on control barrier...
ICRA'24 Control-Barrier-Aided Teleoperation with VI-SLAM for Safe MAV Navigation (Presentation)
มุมมอง 2453 หลายเดือนก่อน
This is a short presentation for the ICRA 2024 work titled "Control-Barrier-Aided Teleoperation with Visual-Inertial SLAM for Safe MAV Navigation in Complex Environments." Researchers: Siqi Zhou, Sotiris Papatheodorou, Stefan Leutenegger, and Angela P. Schoellig Arxiv Preprint: arxiv.org/abs/2403.04331 Abstract: In this paper, we consider a Micro Aerial Vehicle (MAV) system teleoperated by a no...
Welcome to the Learning Systems and Robotics Lab (2024)
มุมมอง 2854 หลายเดือนก่อน
The video features demonstrations of our lab as of June 2024. If you come to our lab, this is what you can see! The demonstrations include: * Interactive wand demo * Periodic swarm patterns for quadrotors in synchronization with music * Online multiagent motion planning with distributed model predictive control * An alternating minimization approach for multiagent motion planning in cluttered e...
Force Push: Robust Single-Point Pushing with Force Feedback
มุมมอง 2584 หลายเดือนก่อน
This video accompanies the paper "Force Push: Robust Single-Point Pushing with Force Feedback", published in IEEE Robotics and Automation Letters (RA-L). Here we demonstrate quasistatic robotic planar pushing with single-point contact using only force feedback to sense the pushed object. Researchers: Adam Heins and Angela P. Schoellig Paper: doi.org/10.1109/LRA.2024.3414180 Arxiv: arxiv.org/abs...
ACC24: Practical Considerations for Implementing Control Barrier Function-Based Safety Filters
มุมมอง 2695 หลายเดือนก่อน
Title: Practical Considerations for Discrete-Time Implementations of Continuous-Time Control Barrier Function-Based Safety Filters Abstract: Safety filters based on control barrier functions (CBFs) have become a popular method to guarantee safety for uncertified control policies, e.g., as resulting from reinforcement learning. Here, safety is defined as staying in a pre-defined set, the safe se...
ICRA'24 Control-Barrier-Aided Teleoperation with VI-SLAM for Safe MAV Navigation
มุมมอง 2746 หลายเดือนก่อน
This video presents the experimental results of the ICRA 2024 work titled "Control-Barrier-Aided Teleoperation with Visual-Inertial SLAM for Safe MAV Navigation in Complex Environments." Researchers: Siqi Zhou, Sotiris Papatheodorou, Stefan Leutenegger, and Angela P. Schoellig Arxiv Preprint: arxiv.org/abs/2403.04331 Abstract: In this paper, we consider a Micro Aerial Vehicle (MAV) system teleo...
Optimized Control Invariance Conditions for Uncertain Input-Constrained Nonlinear Control Systems
มุมมอง 3288 หลายเดือนก่อน
Published in L-CSS 2023 and to be presented at ACC 2024 Title: Optimized Control Invariance Conditions for Uncertain Input-Constrained Nonlinear Control Systems Abstract: Providing safety guarantees for learning-based controllers is important for real-world applications. One approach to realizing safety for arbitrary control policies is safety filtering. If necessary, the filter modifies contro...
LSY Invited Talks: Kimberly McGuire "Swarming with the Lighthouse Positioning System and ROS2"
มุมมอง 1439 หลายเดือนก่อน
Invited Speaker: Kimberly McGuire (knmcguire.github.io/ln/), Bitcraze Title: "Swarming with the Lighthouse Positioning System and ROS2" Date of the Talk: 06.04.2023 Abstract: This talk will be about the Crazyflie, ROS2, and the lighthouse positioning system. Kimberly will start the talk with an introduction to Bitcraze and an overview of Crazyflie for those who are not familiar with the platfor...
LSY Invited Talks: Oier Mees "Interactive Language Instructable Robot Learning"
มุมมอง 1229 หลายเดือนก่อน
Invited Speaker: Oier Mees (www.oiermees.com/), Freiburg University Title: "Interactive Language Instructable Robot Learning" Date of the Talk: 13.06.2023 Abstract: Despite considerable progress in robot learning, and contrary to the general public's expectations, the vast majority of robots deployed in the real world today remain restricted to a narrow set of preprogrammed behaviors for specif...
CDC 2023 Workshop on Open-Source Code
มุมมอง 57810 หลายเดือนก่อน
Webpage: www.dynsyslab.org/cdc-2023-workshop-on-benchmarking-reproducibility-and-open-source-code-in-controls/ Over the past years, the scientific community has grown more cognizant of the importance and challenges of transparent and reproducible research. This topic has become increasingly important given the rise of complex algorithms (e.g., machine learning models or optimization-based algor...
CDC23: Multi-Step Model Predictive Safety Filters: Reducing Chattering
มุมมอง 298ปีที่แล้ว
Published and presented at CDC 2023 Title: Multi-Step Model Predictive Safety Filters: Reducing Chattering by Increasing the Prediction Horizon Abstract: Learning-based controllers have demonstrated superior performance compared to classical controllers in various tasks. However, providing safety guarantees is not trivial. Safety, the satisfaction of state and input constraints, can be guarante...
Keep it Upright: Nonprehensile Object Transportation on a Mobile Manipulator
มุมมอง 4.7Kปีที่แล้ว
Keep it Upright: Nonprehensile Object Transportation on a Mobile Manipulator
Finding the Right Place: Sensor Placement for UWB TDOA Localization in Cluttered Indoor Environments
มุมมอง 695ปีที่แล้ว
Finding the Right Place: Sensor Placement for UWB TDOA Localization in Cluttered Indoor Environments
AMSwarm: An Alternating Minimization Approach for Safe Motion Planning of Quadrotor Swarms
มุมมอง 1.5Kปีที่แล้ว
AMSwarm: An Alternating Minimization Approach for Safe Motion Planning of Quadrotor Swarms
IROS 2022 Safe Robot Learning Competition Awards
มุมมอง 849ปีที่แล้ว
IROS 2022 Safe Robot Learning Competition Awards
IROS 2022 Keynote: Safe Learning in Robotics by Prof. Angela Schoellig
มุมมอง 3Kปีที่แล้ว
IROS 2022 Keynote: Safe Learning in Robotics by Prof. Angela Schoellig
IROS 2022 Safe Robot Learning Competition Winning Teams
มุมมอง 315ปีที่แล้ว
IROS 2022 Safe Robot Learning Competition Winning Teams
IROS 2022 Safe Robot Learning Competition
มุมมอง 7312 ปีที่แล้ว
IROS 2022 Safe Robot Learning Competition
SiQi Zhou Doctoral Seminar: Neural Networks as Add-on Modules for Improving Robot Performance
มุมมอง 6902 ปีที่แล้ว
SiQi Zhou Doctoral Seminar: Neural Networks as Add-on Modules for Improving Robot Performance
Doctoral Seminar: Flying Flat Out - Fast Multirotor Flight in Real-World Environments
มุมมอง 7502 ปีที่แล้ว
Doctoral Seminar: Flying Flat Out - Fast Multirotor Flight in Real-World Environments
Stochastic modeling of tag installation error for robust on-manifold tag-based localization
มุมมอง 832 ปีที่แล้ว
Stochastic modeling of tag installation error for robust on-manifold tag-based localization
ICRA 2022 Workshop Releasing Robotics into the Wild: Animesh Garg Talk
มุมมอง 1962 ปีที่แล้ว
ICRA 2022 Workshop Releasing Robotics into the Wild: Animesh Garg Talk
ICRA 2022 Workshop Releasing Robotics into the Wild: David Hsu Talk
มุมมอง 1652 ปีที่แล้ว
ICRA 2022 Workshop Releasing Robotics into the Wild: David Hsu Talk
ICRA 2022 Workshop Releasing Robotics into the Wild: Davide Scaramuzza Talk
มุมมอง 1502 ปีที่แล้ว
ICRA 2022 Workshop Releasing Robotics into the Wild: Davide Scaramuzza Talk
ICRA 2022 Workshop Releasing Robotics into the Wild: Erwin Coumans
มุมมอง 1982 ปีที่แล้ว
ICRA 2022 Workshop Releasing Robotics into the Wild: Erwin Coumans
ICRA 2022 Workshop Releasing Robotics into the Wild: Ingmar Posner
มุมมอง 2522 ปีที่แล้ว
ICRA 2022 Workshop Releasing Robotics into the Wild: Ingmar Posner
amazing
After watching your video, I found it very interesting, may I ask which way do you describe the OTG of the present pose, Euler Angle, quaternion or SO(3), or something else?
Amazing....
Good work but I have access to datasets
I agree. Got the same results
great work!
Nice Research Topic! Really looking forward for the paper, keep going with the good work!
This is awesome
Are you using reverse TDoA?
Amazing!
This is an amazing work @DR. Melissa Greeff in the domain of Unmanned Aerial Vehicles.
PЯӨMӨƧM 🏃
Looks pretty good! Is the omnidirectional mobile robot?
Good job!
how did you do it can you share with me , thank you
Thank you for sharing.
真不错
Thank you Adrian
Amazing ML research.
Great work! Would be amazing if there's force-torque sensor in end-effector Also feedback Cartesian impedance control can be implemented to track a straight line
Hi, the session is nice what will be the Targets for the GP model (During Training), can you please tell me about the prior model and unknown model, is it like train and test respectively. Hope for a reply, Thanks in Advance
Great Work
Thanks for very clear explanation.
Hi, the session is nice what will be the Targets for the GP model (During Training), can you please tell me about the prior model and unknown model, is it like train and test respectively. Hope for a reply, Thanks in Advance
kawaii crazyflie2.0 lol
I am joining this lab one day. Hopefully soon.
wow,thats amazing ! But I wonder how could you get the coordinate of 3d points in google earth?
Amazing work . Did you tried this algorithm with trajectory planning including analiticly based inverse kinematic solution instead of QP ?
I need help regarding this simulation. Thanks
Make robots