Autonomous Inverted Helicopter Flight via Reinforcement Learning

Ng, Andrew Y.; Coates, Adam; Diel, Mark; Ganapathi, Varun; Schulte, Jamie; Tse, Ben; Berger, Eric; Liang, Eric

doi:10.1007/11552246_35

Cited by 427 publications

(327 citation statements)

References 5 publications

Supporting

Mentioning

324

Contrasting

Unclassified

Order By: Relevance

“…Several different definitions of the MDP exist. In the remainder of this thesis, we will use the following one from [44]. An MDP can be represented as a tuple < S, A, D, P (s t , a t ), γ, ρ(s t , a t ) >, consisting of:…”

Section: The Markov Decision Processmentioning

confidence: 99%

“…After the off-line training, they expose the controller on-line to more demanding flight maneuvers. Ng uses a actor-only method to fly an autonomous model helicopter, by estimating the cost value from Monte-Carlo roll-outs [44] (see subsection 2.5.3).…”

Section: Temporal-difference Learning: Actors Critics Actor-criticsmentioning

confidence: 99%

“…For one, it is able to learn to control a variety of systems, so for a slightly different system, ones does not need to fully redesign a new controller. Furthermore, it can handle extremely non-linear and complex behavior, such as controlling helicopters [44], and autonomous cars [27] and playing video games such as Doom [28] and Super Mario [67]. RL agents are able to keep learning while in operation, and therefore have the potential to adapt to new circumstances.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Flexible Heuristic Dynamic Programming for Reinforcement Learning in Quad-Rotors

Helmer

Visser

Kampen

2018

2018 AIAA Information Systems-Aiaa Infotech @ Aerospace

View full text Add to dashboard Cite

Section: The Markov Decision Processmentioning

confidence: 99%

Section: Temporal-difference Learning: Actors Critics Actor-criticsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Flexible Heuristic Dynamic Programming for Reinforcement Learning in Quad-Rotors

Helmer

Visser

Kampen

2018

2018 AIAA Information Systems-Aiaa Infotech @ Aerospace

View full text Add to dashboard Cite

“…Since the recorded data maps directly to the learner platform, this demonstration technique best minimizes the introduction of correspondence issues into an LfD system. Examples of successful teleoperated LfD systems include both real [10] and simulated [7] robot applications.…”

Section: Learning From Demonstrationmentioning

confidence: 99%

Learning Mobile Robot Motion Control from Demonstrated Primitives and Human Feedback

Argall

Browning

Veloso

2011

Springer Tracts in Advanced Robotics

View full text Add to dashboard Cite

Task demonstration is one effective technique for developing robot motion control policies. As tasks become more complex, however, demonstration can become more difficult. In this work we introduce a technique that uses corrective human feedback to build a policy able to perform an undemonstrated task from simpler policies learned from demonstration. Our algorithm first evaluates and corrects the execution of motion primitive policies learned from demonstration. The algorithm next corrects and enables the execution of a larger task built from these primitives. Within a simulated robot motion control domain, we validate that a policy for an undemonstrated task is successfully built from motion primitives learned from demonstration under our approach. We show feedback to both aid and enable policy development, improving policy performance in success, speed and efficiency.

show abstract

“…RL has been used to control a model helicopter than can hover while inverted in air [7]. Other ML techniques have been applied to directly control simulated bipedal robots: in [8] a central pattern generator was used for rhythm generation in the hips and knees of a simulated bipedal robot, and a dynamics controller was used to control the ankles of robot.…”

Section: Related Workmentioning

confidence: 99%

Autonomous Learning of Stable Quadruped Locomotion

Saggar¹,

D'Silva

Kohl³

et al. 2007

RoboCup 2006: Robot Soccer World Cup X

View full text Add to dashboard Cite

Abstract. A fast gait is an essential component of any successful team in the RoboCup 4-legged league. However, quickly moving quadruped robots, including those with learned gaits, often move in such a way so as to cause unsteady camera motions which degrade the robot's visual capabilities. This paper presents an implementation of the policy gradient machine learning algorithm that searches for a parameterized walk while optimizing for both speed and stability. To the best of our knowledge, previous learned walks have all focused exclusively on speed. Our method is fully implemented and tested on the Sony Aibo ERS-7 robot platform. The resulting gait is reasonably fast and considerably more stable compared to our previous fast gaits. We demonstrate that this stability can significantly improve the robot's visual object recognition.

show abstract

Autonomous Inverted Helicopter Flight via Reinforcement Learning

Cited by 427 publications

References 5 publications

Flexible Heuristic Dynamic Programming for Reinforcement Learning in Quad-Rotors

Flexible Heuristic Dynamic Programming for Reinforcement Learning in Quad-Rotors

Learning Mobile Robot Motion Control from Demonstrated Primitives and Human Feedback

Autonomous Learning of Stable Quadruped Locomotion

Contact Info

Product

Resources

About