- 56
- 256 826
MIT AERA
Приєднався 27 бер 2011
Sertac Karaman's Research Group at Massachusetts Institute of Technology
Multi-Fidelity Reinforcement Learning for Time-Optimal Quadrotor Re-planning
This video demonstrates the experiments described in our recent paper entitled "Multi-Fidelity Reinforcement Learning for Time-Optimal Quadrotor Re-planning."
Arxiv: arxiv.org/abs/2403.08152
Github: github.com/mit-aera/mfrlTrajectory
High-speed online trajectory planning for UAVs poses a significant challenge due to the need for precise modeling of complex dynamics while also being constrained by computational limitations. This paper presents a multi-fidelity reinforcement learning method (MFRL) that aims to effectively create a realistic dynamics model and simultaneously train a planning policy that can be readily deployed in real-time applications. The proposed method involves the co-training of a planning policy and a reward estimator; the latter predicts the performance of the policy's output and is trained efficiently through multi-fidelity Bayesian optimization. This optimization approach models the correlation between different fidelity levels, thereby constructing a high-fidelity model based on a low-fidelity foundation, which enables the accurate development of the reward model with limited high-fidelity experiments. The framework is further extended to include real-world flight experiments in reinforcement learning training, allowing the reward model to precisely reflect real-world constraints and broadening the policy's applicability to real-world scenarios. We present rigorous evaluations by training and testing the planning policy in both simulated and real-world environments. The resulting trained policy not only generates faster and more reliable trajectories compared to the baseline snap minimization method, but it also achieves trajectory updates in 2 ms on average, while the baseline method takes several minutes.
Arxiv: arxiv.org/abs/2403.08152
Github: github.com/mit-aera/mfrlTrajectory
High-speed online trajectory planning for UAVs poses a significant challenge due to the need for precise modeling of complex dynamics while also being constrained by computational limitations. This paper presents a multi-fidelity reinforcement learning method (MFRL) that aims to effectively create a realistic dynamics model and simultaneously train a planning policy that can be readily deployed in real-time applications. The proposed method involves the co-training of a planning policy and a reward estimator; the latter predicts the performance of the policy's output and is trained efficiently through multi-fidelity Bayesian optimization. This optimization approach models the correlation between different fidelity levels, thereby constructing a high-fidelity model based on a low-fidelity foundation, which enables the accurate development of the reward model with limited high-fidelity experiments. The framework is further extended to include real-world flight experiments in reinforcement learning training, allowing the reward model to precisely reflect real-world constraints and broadening the policy's applicability to real-world scenarios. We present rigorous evaluations by training and testing the planning policy in both simulated and real-world environments. The resulting trained policy not only generates faster and more reliable trajectories compared to the baseline snap minimization method, but it also achieves trajectory updates in 2 ms on average, while the baseline method takes several minutes.
Переглядів: 422
Відео
WiSwarm: Age-of-Information-based Wireless Networking for Collaborative Teams of UAVs
Переглядів 1,1 тис.Рік тому
This video showcases experiments for our recent paper entitled "WiSwarm: Age-of-Information-based Wireless Networking for Collaborative Teams of UAVs". The paper will be presented at IEEE INFOCOM 2023 and is available at arxiv.org/abs/2212.03298. The Age-of-Information (AoI) metric has been widely studied in the theoretical communication networks and queuing systems literature. However, experim...
Real-Time Generation of Time-Optimal Quadrotor Trajectories with Semi-Supervised Seq2Seq Learning
Переглядів 554Рік тому
This video demonstrates the experiments described in our recent publication entitled "Real-Time Generation of Time-Optimal Quadrotor Trajectories with Semi-Supervised Seq2Seq Learning". The paper will be presented at the Conference on Robot Learning (CoRL) 2022. The paper is available at openreview.net/forum?id=ZJYbW8Cwys Generating time-optimal quadrotor trajectories is challenging due to the ...
Global Incremental Flight Control for Agile Maneuvering of a Tailsitter Flying Wing
Переглядів 1,4 тис.Рік тому
This paper proposes a novel control law for accurate tracking of agile trajectories using a tailsitter flying wing unmanned aerial vehicle (UAV) that transitions between vertical take-off and landing (VTOL) and forward flight. The global control formulation enables maneuvering throughout the flight envelope, including uncoordinated flight with sideslip. Differential flatness of the nonlinear ta...
Aerobatic Trajectory Generation for a VTOL Fixed-Wing Aircraft Using Differential Flatness
Переглядів 21 тис.2 роки тому
Paper available at: arxiv.org/abs/2207.03524 Project website: aera.mit.edu/projects/TailsitterAerobatics This paper proposes a novel algorithm for aerobatic trajectory generation for a vertical take-off and landing (VTOL) tailsitter flying wing aircraft. The algorithm differs from existing approaches for fixed-wing trajectory generation, as it considers a realistic six-degree-of-freedom (6DOF) ...
Aerobatic Sequence for Three VTOL Fixed-Wing Aircraft
Переглядів 4,2 тис.2 роки тому
Paper available at: arxiv.org/abs/2207.03524 Project website: aera.mit.edu/projects/TailsitterAerobatics This paper proposes a novel algorithm for aerobatic trajectory generation for a vertical take-off and landing (VTOL) tailsitter flying wing aircraft. The algorithm differs from existing approaches for fixed-wing trajectory generation, as it considers a realistic six-degree-of-freedom (6DOF) ...
Cooperative Multi-Agent Trajectory Generation with Modular Bayesian Optimization
Переглядів 7202 роки тому
This video showcases experiments for our recent paper entitled "Cooperative Multi-Agent Trajectory Generation with Modular Bayesian Optimization". The paper will be presented at Robotics: Science and Systems (RSS) 2022. The paper is available at arxiv.org/abs/2206.00726 We present a modular Bayesian optimization framework that efficiently generates time-optimal trajectories for a cooperative mu...
Global Trajectory-tracking Control for a Tailsitter Flying Wing in Agile Uncoordinated Flight
Переглядів 1 тис.3 роки тому
We propose a novel control law for accurate tracking of agile trajectories using a tailsitter flying wing micro unmanned aerial vehicle (UAV) that transitions between vertical take-off and landing (VTOL) and forward flight. Our global control formulation enables agile maneuvering throughout the flight envelope, including uncoordinated flight conditions with lateral accelerations and sideslip. W...
Multi-Fidelity Black-Box Optimization for Time-Optimal Quadrotor Maneuvers
Переглядів 5 тис.3 роки тому
This video showcases experiments for our recent paper entitled "Multi-Fidelity Black-Box Optimization for Time-Optimal Quadrotor Maneuvers". The paper is available at doi.org/10.1177/02783649211033317. We consider the problem of generating a time-optimal quadrotor trajectory for highly maneuverable vehicles, such as quadrotor aircraft. The problem is challenging because the optimal trajectory i...
Multi-Modal Motion Planning Using Composite Pose Graph Optimization
Переглядів 5563 роки тому
This video showcases experiments for our recent paper entitled "Multi-Modal Motion Planning Using Composite Pose Graph Optimization". The paper will be presented at the 2021 International Conference on Robotics and Automation (ICRA 2021) and is available at arxiv.org/abs/2107.02384. We present a motion planning framework for multi-modal vehicle dynamics. Our approach employs transcription of t...
FlightGoggles: A Modular Framework for Photorealistic Simulation
Переглядів 3,1 тис.3 роки тому
FlightGoggles is a photorealistic sensor simulator for perception-driven robotic vehicles. The key contributions of FlightGoggles are twofold. First, FlightGoggles provides photorealistic exteroceptive sensor simulation using graphics assets generated with photogrammetry. Second, it also provides the ability to combine (i) synthetic exteroceptive measurements generated in silico in real time an...
FlightGoggles Arcade teaser
Переглядів 3893 роки тому
FlightGoggles Arcade is a drone racing game using the FlightGoggles Stata Center Environment. This video is a teaser footage, showing the basic elements of game play. Coming soon!
FlightGoggles Stata Center Environment
Переглядів 5903 роки тому
A walkthrough of the Stata Center environment in FlightGoggles, showcasing some of the prominent graphics assets. The environment is a digital replica of two floors of MIT's Stata Center - a Frank-Gehry-designed landmark of Boston.
Accurate Tracking of Aggressive Quadrotor Trajectories using INDI and Differential Flatness
Переглядів 2,1 тис.4 роки тому
This video showcases experiments for our recent paper entitled "Accurate Tracking of Aggressive Quadrotor Trajectories using Incremental Nonlinear Dynamic Inversion and Differential Flatness." The paper is available at arxiv.org/abs/1809.04048 In the paper, we propose a novel control law for accurate tracking of aggressive quadcopter trajectories. The proposed method tracks position and yaw ang...
Multi-Fidelity Black-Box Optimization for Time-Optimal Quadrotor Maneuvers
Переглядів 9744 роки тому
This video showcases experiments for our recent paper entitled "Multi-Fidelity Black-Box Optimization for Time-Optimal Quadrotor Maneuvers". The paper will be presented at Robotics: Science and Systems (RSS) 2020. The paper is available here: arxiv.org/abs/2006.02513. In the paper, we consider the problem of generating a time-optimal quadrotor trajectory that attains a set of prescribed waypoin...
AlphaPilot FlightGoggles simulation challenge: Finish Line
Переглядів 5555 років тому
AlphaPilot FlightGoggles simulation challenge: Finish Line
AlphaPilot FlightGoggles simulation challenge: Gate 14 view
Переглядів 4135 років тому
AlphaPilot FlightGoggles simulation challenge: Gate 14 view
AlphaPilot FlightGoggles simulation challenge: Center of the course view
Переглядів 1665 років тому
AlphaPilot FlightGoggles simulation challenge: Center of the course view
AlphaPilot FlightGoggles simulation challenge: Gate 13 view
Переглядів 1845 років тому
AlphaPilot FlightGoggles simulation challenge: Gate 13 view
AlphaPilot FlightGoggles simulation challenge: Gate 21 view 2
Переглядів 1375 років тому
AlphaPilot FlightGoggles simulation challenge: Gate 21 view 2
AlphaPilot FlightGoggles simulation challenge: Gate 21 view 1
Переглядів 2765 років тому
AlphaPilot FlightGoggles simulation challenge: Gate 21 view 1
AlphaPilot FlightGoggles simulation challenge: Tile view (with trails)
Переглядів 2175 років тому
AlphaPilot FlightGoggles simulation challenge: Tile view (with trails)
AlphaPilot FlightGoggles simulation challenge: Tile view (no trails)
Переглядів 3305 років тому
AlphaPilot FlightGoggles simulation challenge: Tile view (no trails)
AlphaPilot FlightGoggles simulation challenge: Overhead view
Переглядів 5535 років тому
AlphaPilot FlightGoggles simulation challenge: Overhead view
FlightGoggles Abandoned Factory Environment
Переглядів 1,2 тис.5 років тому
FlightGoggles Abandoned Factory Environment
FlightGoggles in AlphaPilot simulation qualifiers - Teaser video
Переглядів 2,8 тис.5 років тому
FlightGoggles in AlphaPilot simulation qualifiers - Teaser video
FlightGoggles in AlphaPilot simulation qualifiers
Переглядів 1,8 тис.5 років тому
FlightGoggles in AlphaPilot simulation qualifiers
FlightGoggles: Photorealistic Sensor Simulation for Perception-driven Robotics
Переглядів 3,6 тис.5 років тому
FlightGoggles: Photorealistic Sensor Simulation for Perception-driven Robotics
Accurate Tracking of Aggressive Quadrotor Trajectories
Переглядів 4 тис.6 років тому
Accurate Tracking of Aggressive Quadrotor Trajectories
I need simulation model of tailsitter vtol uav?
I need simulation model of tailsitter vtol uav?
Outstanding !
Excellent work! I would be grateful if you could tell what effect did you use to visualize the motion of the trajectory 2:05 in repeated drone motion?
Quality work, thank you, What simulator did you use? What tools for plotting and visualization?
Wonderful work, I cant quite see are the motors also articulated. I feel they must be, huge control surfaces! They are moving in such a fluid manner that it looks easy, which of course its not! Delightful
motors are fixedd
Wow, impressive! I hope the next step isn't skynet...
could agree with some commenters below - wing is orphan element here, these drones don't need them for flight in at least these modes. would be interesting to know if these wings could even allow to plane safely with switch off motors.
3d positioning and handling of all the complex dynamics is extremely impressive, but are these aircraft really 'fixed-wing'? really? isn't this just a slightly worse version of quad copters with extra steps?
No, it’s not a worse quadcopter with extra steps.
The drones are not really using the wings for lifting the device in horizontal flight ?
hello, what is your drone? where did you buy it?
No sound
It's incredible, but it feels like kind of a stretch to call it a fixed-wing aircraft. When it's mostly a 2 propeller drone with a fuselage that is incidentally shaped like a wing, but never really works like one too much and most demonstrated flight modes probably don't offer a safe transition to a glide in case the motors die.
No
Cool UFO
This is fascinating! Great work!
This was pretty interesting, thank you UA-cam algorithm
Are the graphs presented just a plan or an actual location?
Great research, Have you considered same hight ( Z ) for each drone ( or all same) to make random new trajectory?
can you share this project s matlab codes ? how can i reach you
awesome
Could you tell me what optical encoder model you used? I wonder if you used a motor equipped with an optical encoder. Thanks :)
COOOL!
Simply amazing!
what drone are you guys using? great projects btw
Impressive!
Amazing work!
Awesome work. Which references (text books or articles) can read to achieve this kind of work? Thank you
Tebrikler hocam harika bir çalışma.
I didnt understand anything but it all looks very cool.
Hi, wich model is that drone? Is it the same as this? : articulo.mercadolibre.com.mx/MLM-655610994-parrot-mini-drone-rolling-spider-aereo-y-terrestre-_JM
Is it possible to gain access to the matlab code for the line follower projects?
Is it possible to have information on the line following drone? I am need of guidance for my own project
Hi, I would like to pass your course remotely, but unfortunately the links on the site are non-working. Could you give me access to the training materials? ouafomandela@gmail.com. Thank
Great! Congratulations to the professor and students!
Great stuff! It would be nice to know which drone belongs to which team.
Hi! I really love this video! I was curious about the self-driving algorithm and sensor data used. Do they use only images from the camera to navigate? Do they use data from the lidar? Or map? Thanks!
Opensource the code or it didn't happen. 😜
Does it only work for 3 seconds at a time? More of the video is black than preview.
Hello Sir, Can you please share the base paper or documentation for the flight goggles?
nice!. can't wait to try it out.
Impressive
Hocam Merhabalar,Kolay gelsin. Bu alan üzerinde benim de ilgim var. Literatür eksiğim çok sizce nereden başlamalıyım? Kaynak önerebilir misiniz? Şimdiden teşekkür ediyorum. Başarılarına başarı katma dileğiyle...
nice work. direct feedback control of motor RPMs has been missing from quadrotor hardware for way too long.
Amazing Sertac, congratulations.
Hi, I would like to get your course from the provided links fast.scripts.mit.edu/dronecontrol/ but unfortunately the links are borken. Could you give me access to the training materials to my email m.belkheiri@lagh-univ.dz or mbelkhiri@yahoo.com
Hi sir, I need more information, documents and lab handouts on the rolling spider. Your work is very interesting, could you plz fix the lab handout links on the website fast.scripts.mit.edu/dronecontrol/labs/ or send it to me mbelkhiri@yahoo.com Regards
It is a great work could you please send me a link to the lab handouts. Regards
Hi, what type of Bluetooth are you using for connection? I am having trouble connecting the RS to the host PC after following every instructions on both linux and windows machines...
Hi, I would like to pass your course remotely, but unfortunately the links on the site are non-working. Could you give me access to the training materials?