DOCSLIB.ORG

Reinforcement Learning Lecture Inverse Reinforcement Learning

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Reinforcement Learning Lecture Inverse Reinforcement Learning Reinforcement Learning Inverse Reinforcement Learning Inverse RL, behaviour cloning, apprenticeship learning, imitation learning. Vien Ngo MLR, University of Stuttgart Outline Introduction to Inverse RL • Inverse RL vs. behavioral cloning • IRL algorithms • (Inspired from a lecture from Pieter Abbeel.) 2/?? Inverse RL: Informal Definition Given Measurements of an agent’s behaviour π over time (s ; a ; s0 ), in • t t t different circumstances. If possible, given transition model (not given reward function). π Goal: Find the reward function R (s; a; s0). • 3/?? Inverse Reinforcement Learning Advanced Robotics Vien Ngo Machine Learning & Robotics lab, University of Stuttgart Universittsstrae 38, 70569 Stuttgart, Germany June 17, 2014 1 A Small Maze Domain Given a 2 4 maze as in figure. Given a MDP/R = S,A,P , where S is a state space consisting of 8 states; A is an action × { } space consisting of four movement actions move-left,move-right,move-down,move-up , the optimal policy (arrows in { } figure), and the transition function T (s, a, s0) = P (s0 s, a). | Goal π∗ 1 Given γ = 0.95, compute (I γP )− . • − Then, write full formulation of the LP problem in slide # 12. • Bonus: use fminimax function in Matlab (or other LP library) to find R∗. • 4/?? 1 inspired from a poster of Boularias, Kober, Peters. Construction of an intelligent agent in a particular domain: Car driver, • helicopter (Ng et al),... (imitation learning, apprenticeship learning) Motivation: Two Sources The potential use of RL/related methods as computational model for • animal and human learning: bee foraging (Montague et al 1995), song-bird vocalization (Doya & Sejnowski 1995), ... 5/?? Motivation: Two Sources The potential use of RL/related methods as computational model for • animal and human learning: bee foraging (Montague et al 1995), song-bird vocalization (Doya & Sejnowski 1995), ... Construction of an intelligent agent in a particular domain: Car driver, • helicopter (Ng et al),... (imitation learning, apprenticeship learning) 5/?? Examples Car driving simulation • Abbeel et al 2004, etc. Autonomous Helicopter Flight • Andrew Ng et. al. Urban navigation • Ziebart, Maas, Bagnell and Dey, AAAI 2008 (route recommendation, and destination prediction) etc. • 6/?? Problem Formulation Given • – State space , action space cA. S – Transition model T (s; a; s0) = P (s0 s; a) j – not given reward function R(s; a; s0). – Teacher’s demonstration (from teacher’s policy π∗): s0; a0; s1; a1;:::; IRL: • – Recover R. Apprenticeship learning via IRL • – Use R to compute a good policy. Behaviour cloning: • – Using supersived-learning to learn the teacher’s policy. 7/?? IRL vs. behavioral cloning 8/?? Behavioral cloning: can only mimic the trajectory of the teacher, then • can not: with change of goal/destination, and non-Markovian environment (e.g. car driving). IRL vs. Behavioral cloning is R^∗ vs. π^∗. • IRL vs. Behavioral cloning Behavioral cloning: Formulated as a supervised-learning problem. • (Using SVM, Neural networks, deep learning,...) – Given (s0; a0); (s1; a1);:::; generated from a policy π∗. – Estimate a policy mapping s to a. 9/?? IRL vs. Behavioral cloning is R^∗ vs. π^∗. • IRL vs. Behavioral cloning Behavioral cloning: Formulated as a supervised-learning problem. • (Using SVM, Neural networks, deep learning,...) – Given (s0; a0); (s1; a1);:::; generated from a policy π∗. – Estimate a policy mapping s to a. Behavioral cloning: can only mimic the trajectory of the teacher, then • can not: with change of goal/destination, and non-Markovian environment (e.g. car driving). 9/?? IRL vs. Behavioral cloning Behavioral cloning: Formulated as a supervised-learning problem. • (Using SVM, Neural networks, deep learning,...) – Given (s0; a0); (s1; a1);:::; generated from a policy π∗. – Estimate a policy mapping s to a. Behavioral cloning: can only mimic the trajectory of the teacher, then • can not: with change of goal/destination, and non-Markovian environment (e.g. car driving). IRL vs. Behavioral cloning is R^∗ vs. π^∗. • 9/?? Inverse Reinforcement Learning 10/?? Challenges? • – R = 0 is a solution (rewrad function ambiguity), and multiple R∗ satisfy the above condition. – π∗ is only given partially through trajectories, then how to evaluate the expectation terms. – must assume the expert is optimal – R.H.S is computationally expensive, i.e enumerate all policies IRL: Mathematical Formulation Given • – State space , action space cA. S – Transition model T (s; a; s0) = P (s0 s; a) j – not given reward function R(s; a; s0). – Teacher’s demonstration (from teacher’s policy π∗): s0; a0; s1; a1;:::; Find R∗, such that • h X1 t i h X1 t i E γ R∗(s ) π∗ E γ R∗(s ) π ; π t j ≥ t j 8 t=0 t=0 11/?? IRL: Mathematical Formulation Given • – State space , action space cA. S – Transition model T (s; a; s0) = P (s0 s; a) j – not given reward function R(s; a; s0). – Teacher’s demonstration (from teacher’s policy π∗): s0; a0; s1; a1;:::; Find R∗, such that • h X1 t i h X1 t i E γ R∗(s ) π∗ E γ R∗(s ) π ; π t j ≥ t j 8 t=0 t=0 Challenges? • – R = 0 is a solution (rewrad function ambiguity), and multiple R∗ satisfy the above condition. – π∗ is only given partially through trajectories, then how to evaluate the expectation terms. – must assume the expert is optimal – R.H.S is computationally expensive, i.e enumerate all policies 11/?? IRL as linear programming with l • 1 S j j n o X a∗ b a∗ 1 max min (P (i) P (i))(I γP )− R λ R 1 b =a − − − jj jj i=1 2A ∗ s.t. a∗ b a∗ 1 (P P )(I γP )− R 0 − − ≥ R(i) R ≤ max – Maximize the sum of differences between the values of the optimal action and the next-best. – With l1 penalty. IRL: Finite state spaces Bellman equations • π π 1 V = (I γP )− R − Then IRL finds R such that • a∗ a a∗ 1 (P P )(I γP )− R 0; a − − ≥ 8 (if consider only deterministic policies) 12/?? IRL: Finite state spaces Bellman equations • π π 1 V = (I γP )− R − Then IRL finds R such that • a∗ a a∗ 1 (P P )(I γP )− R 0; a − − ≥ 8 (if consider only deterministic policies) IRL as linear programming with l • 1 S j j n o X a∗ b a∗ 1 max min (P (i) P (i))(I γP )− R λ R 1 b =a − − − jj jj i=1 2A ∗ s.t. a∗ b a∗ 1 (P P )(I γP )− R 0 − − ≥ R(i) R ≤ max – Maximize the sum of differences between the values of the optimal action and the next-best. – With l1 penalty. 12/?? The optimization problem: finding w∗ such that • w∗>.η(π∗) w∗>.η(π) ≥ η(π) can be evaluated with sampled trajectories from π. • N T 1 X Xi η(π) = γtφ(s ) N t i=1 t=0 IRL: With FA in large state spaces n Using FA: R(s) = w>.φ(s), where w R , and φ : R. • 2 S 7! Thus, • h t i h t i E γ R(s ) π = E γ w>φ(s ) π t j t j h t i = w>E γ φ(s ) π = w>.η(π) t j 13/?? IRL: With FA in large state spaces n Using FA: R(s) = w>.φ(s), where w R , and φ : R. • 2 S 7! Thus, • h t i h t i E γ R(s ) π = E γ w>φ(s ) π t j t j h t i = w>E γ φ(s ) π = w>.η(π) t j The optimization problem: finding w∗ such that • w∗>.η(π∗) w∗>.η(π) ≥ η(π) can be evaluated with sampled trajectories from π. • N T 1 X Xi η(π) = γtφ(s ) N t i=1 t=0 13/?? Apprenticeship learning Abbeel & Ng, 2004 14/?? n 1: Assume R(s) = w>.φ(s), where w R , and φ : R. 2 S 7! 2: Initialize π0 3: for i = 1; 2;::: do 4: Find a reward function such that the teacher maximally outperforms all previously found controllers. max γ γ; w 1 jj jj jj j|≤ s.t. w>.η(π) w>.η(π) + γ; π π0; π1; : : : ; πi 1 ≥ 8 2 f − g 5: Find optimal policy πi for the reward function Rw w.r.t current w. Apprenticeship learning Finding a policy π whose performance is as close to the expert policy’s • performance as possible w∗>.η(π∗) w>.η(π) jj − jj ≤ 15/?? Apprenticeship learning Finding a policy π whose performance is as close to the expert policy’s • performance as possible w∗>.η(π∗) w>.η(π) jj − jj ≤ n 1: Assume R(s) = w>.φ(s), where w R , and φ : R. 2 S 7! 2: Initialize π0 3: for i = 1; 2;::: do 4: Find a reward function such that the teacher maximally outperforms all previously found controllers. max γ γ; w 1 jj jj jj j|≤ s.t. w>.η(π) w>.η(π) + γ; π π0; π1; : : : ; πi 1 ≥ 8 2 f − g 5: Find optimal policy πi for the reward function Rw w.r.t current w15/. ?? Examples 16/?? Simulated Highway Driving Given dynamic model T (s; a; s0) • Each teacher demonstrates 1 minute. • Abbeel et. al. 2004 17/?? Simulated Highway Driving expert demonstration (left), learned control (right) 18/?? Urban Navigation picture from a tutorial of Pieter Abbeel. 19/?? References Andrew Y. Ng, Stuart J. Russell: Algorithms for Inverse Reinforcement Learning. ICML 2000: 663-670 Pieter Abbeel, Andrew Y. Ng: Apprenticeship learning via inverse reinforcement learning. ICML 2004 Pieter Abbeel, Adam Coates, Morgan Quigley, Andrew Y. Ng: An Application of Reinforcement Learning to Aerobatic Helicopter Flight.
Load more

Recommended publications
  • A Review of Robot Learning for Manipulation: Challenges, Representations, and Algorithms
    Journal of Machine Learning Research 22 (2021) 1-82 Submitted 9/19; Revised 8/20; Published 1/21 A Review of Robot Learning for Manipulation: Challenges, Representations, and Algorithms Oliver Kroemer∗ [email protected] School of Computer Science Carnegie Mellon University Pittsburgh, PA 15213, USA Scott Niekum [email protected] Department of Computer Science The University of Texas at Austin Austin, TX 78712, USA George Konidaris [email protected] Department of Computer Science Brown University Providence, RI 02912, USA Editor: Daniel Lee Abstract A key challenge in intelligent robotics is creating robots that are capable of directly in- teracting with the world around them to achieve their goals. The last decade has seen substantial growth in research on the problem of robot manipulation, which aims to exploit the increasing availability of affordable robot arms and grippers to create robots capable of directly interacting with the world to achieve their goals. Learning will be central to such autonomous systems, as the real world contains too much variation for a robot to expect to have an accurate model of its environment, the objects in it, or the skills required to manipulate them, in advance. We aim to survey a representative subset of that research which uses machine learning for manipulation. We describe a formalization of the robot manipulation learning problem that synthesizes existing research into a single coherent framework and highlight the many remaining research opportunities and challenges. Keywords: manipulation, learning, review, robots, MDPs 1. Introduction Robot manipulation is central to achieving the promise of robotics|the very definition of a robot requires that it has actuators, which it can use to effect change on the world.
    [Show full text]
  • Self-Corrective Apprenticeship Learning for Quadrotor Control
    Self-corrective Apprenticeship Learning for Quadrotor Control Haoran Yuan Self-corrective Apprenticeship Learning for Quadrotor Control by Haoran Yuan to obtain the degree of Master of Science at the Delft University of Technology, to be defended publicly on Wednesday December 11, 2019 at 10:00 AM. Student number: 4710118 Project duration: December 12, 2018 – December 11, 2019 Thesis committee: Dr. ir. E. van Kampen, C&S, Daily supervisor Dr. Q. P.Chu, C&S, Chairman of committee Dr. ir. D. Dirkx, A&SM, External examiner B. Sun, MSc, C&S An electronic version of this thesis is available at http://repository.tudelft.nl/. Contents 1 Introduction 1 I Scientific Paper 5 II Literature Review & Preliminary Analysis 23 2 Literature Review 25 2.1 Reinforcement Learning Preliminaries . 25 2.2 Temporal-Difference Learning and Q-Learning . 26 2.2.1 Deep Q Learning . 27 2.3 Policy Gradient Method . 28 2.3.1 Actor-critic and Deep Deterministic Policy Gradient . 30 2.4 Reinforcement Learning for flight Control . 30 2.5 Learning from Demonstrations Preliminaries . 33 2.6 Learning from Demonstrations: Mapping . 34 2.6.1 Mapping Methods without Reinforcement Learning . 35 2.6.2 Mapping Methods with Reinforcement Learning . 35 2.7 Learning from Demonstrations: Planning . 37 2.8 Learning from Demonstrations: Model-Learning . 38 2.8.1 Inverse Reinforcement Learning . 39 2.8.2 Apprenticeship Learning through Inverse Reinforcement Learning . 40 2.8.3 Maximum Entropy Inverse Reinforcement Learning . 42 2.8.4 Relative Entropy Inverse Reinforcement Learning . 42 2.9 Conclusion of Literature Review . 43 3 Preliminary Analysis 45 3.1 Simulation Setup .
    [Show full text]
  • Arxiv:1907.03146V3 [Cs.RO] 6 Nov 2020
    A Review of Robot Learning for Manipulation: Challenges, Representations, and Algorithms Oliver Kroemer* [email protected] School of Computer Science Carnegie Mellon University Pittsburgh, PA 15213, USA Scott Niekum* [email protected] Department of Computer Science The University of Texas at Austin Austin, TX 78712, USA George Konidaris [email protected] Department of Computer Science Brown University Providence, RI 02912, USA Abstract A key challenge in intelligent robotics is creating robots that are capable of directly in- teracting with the world around them to achieve their goals. The last decade has seen substantial growth in research on the problem of robot manipulation, which aims to exploit the increasing availability of affordable robot arms and grippers to create robots capable of directly interacting with the world to achieve their goals. Learning will be central to such autonomous systems, as the real world contains too much variation for a robot to expect to have an accurate model of its environment, the objects in it, or the skills required to manipulate them, in advance. We aim to survey a representative subset of that research which uses machine learning for manipulation. We describe a formalization of the robot manipulation learning problem that synthesizes existing research into a single coherent framework and highlight the many remaining research opportunities and challenges. Keywords: Manipulation, Learning, Review, Robots, MDPs 1. Introduction arXiv:1907.03146v3 [cs.RO] 6 Nov 2020 Robot manipulation is central to achieving the promise of robotics|the very definition of a robot requires that it has actuators, which it can use to effect change on the world.
    [Show full text]
  • Deep Q-Learning from Demonstrations
    The Thirty-Second AAAI Conference on Artificial Intelligence (AAAI-18) Deep Q-Learning from Demonstrations Todd Hester Matej Vecerik Olivier Pietquin Marc Lanctot Google DeepMind Google DeepMind Google DeepMind Google DeepMind [email protected] [email protected] [email protected] [email protected] Tom Schaul Bilal Piot Dan Horgan John Quan Google DeepMind Google DeepMind Google DeepMind Google DeepMind [email protected] [email protected] [email protected] [email protected] Andrew Sendonaris Ian Osband Gabriel Dulac-Arnold John Agapiou Google DeepMind Google DeepMind Google DeepMind Google DeepMind [email protected] [email protected] [email protected] [email protected] Joel Z. Leibo Audrunas Gruslys Google DeepMind Google DeepMind [email protected] [email protected] Abstract clude deep model-free Q-learning for general Atari game- playing (Mnih et al. 2015), end-to-end policy search for con- Deep reinforcement learning (RL) has achieved several high profile successes in difficult decision-making problems. trol of robot motors (Levine et al. 2016), model predictive However, these algorithms typically require a huge amount of control with embeddings (Watter et al. 2015), and strategic data before they reach reasonable performance. In fact, their policies that combined with search led to defeating a top hu- performance during learning can be extremely poor. This may man expert at the game of Go (Silver et al. 2016). An im- be acceptable for a simulator, but it severely limits the appli- portant part of the success of these approaches has been to cability of deep RL to many real-world tasks, where the agent leverage the recent contributions to scalability and perfor- must learn in the real environment.
    [Show full text]
  • Third-Person Imitation Learning
    Published as a conference paper at ICLR 2017 THIRD-PERSON IMITATION LEARNING Bradly C. Stadie1;2, Pieter Abbeel1;3, Ilya Sutskever1 1 OpenAI 2 UC Berkeley, Department of Statistics 3 UC Berkeley, Departments of EECS and ICSI fbstadie, pieter, ilyasu,[email protected] ABSTRACT Reinforcement learning (RL) makes it possible to train agents capable of achiev- ing sophisticated goals in complex and uncertain environments. A key difficulty in reinforcement learning is specifying a reward function for the agent to optimize. Traditionally, imitation learning in RL has been used to overcome this problem. Unfortunately, hitherto imitation learning methods tend to require that demonstra- tions are supplied in the first-person: the agent is provided with a sequence of states and a specification of the actions that it should have taken. While powerful, this kind of imitation learning is limited by the relatively hard problem of collect- ing first-person demonstrations. Humans address this problem by learning from third-person demonstrations: they observe other humans perform tasks, infer the task, and accomplish the same task themselves. In this paper, we present a method for unsupervised third-person imitation learn- ing. Here third-person refers to training an agent to correctly achieve a simple goal in a simple environment when it is provided a demonstration of a teacher achieving the same goal but from a different viewpoint; and unsupervised refers to the fact that the agent receives only these third-person demonstrations, and is not provided a correspondence between teacher states and student states. Our methods primary insight is that recent advances from domain confusion can be utilized to yield domain agnostic features which are crucial during the training process.
    [Show full text]
  • Teaching Agents with Deep Apprenticeship Learning
    Rochester Institute of Technology RIT Scholar Works Theses 6-2017 Teaching Agents with Deep Apprenticeship Learning Amar Bhatt [email protected] Follow this and additional works at: https://scholarworks.rit.edu/theses Recommended Citation Bhatt, Amar, "Teaching Agents with Deep Apprenticeship Learning" (2017). Thesis. Rochester Institute of Technology. Accessed from This Thesis is brought to you for free and open access by RIT Scholar Works. It has been accepted for inclusion in Theses by an authorized administrator of RIT Scholar Works. For more information, please contact [email protected]. Teaching Agents with Deep Apprenticeship Learning by Amar Bhatt A Thesis Submitted in Partial Fulfillment of the Requirements for the Degree of Master of Science in Computer Engineering Supervised by: Assistant Professor Dr. Raymond Ptucha Department of Computer Engineering Kate Gleason College of Engineering Rochester Institute of Technology Rochester, New York June 2017 Approved by: Dr. Raymond Ptucha, Assistant Professor Thesis Advisor, Department of Computer Engineering Dr. Ferat Sahin, Professor Committee Member, Department of Electrical Engineering Dr. Iris Asllani, Assistant Professor Committee Member, Department of Biomedical Engineering Dr. Christopher Kanan, Assistant Professor Committee Member, Department of Imaging Science Louis Beato, Lecturer Committee Member, Department of Computer Engineering Thesis Release Permission Form Rochester Institute of Technology Kate Gleason College of Engineering Title: Teaching Agents with Deep Apprenticeship Learning I, Amar Bhatt, hereby grant permission to the Wallace Memorial Library to reproduce my thesis in whole or part. Amar Bhatt Date iii Dedication This thesis is dedicated to all those who have taught and guided me throughout my years... My loving parents who continue to support me in my academic and personal endeavors.
    [Show full text]
  • Learning Motor Skills: from Algorithms to Robot Experiments
    Learning Motor Skills: From Algorithms to Robot Experiments Erlernen Motorischer Fähigkeiten: Von Algorithmen zu Roboter-Experimenten Zur Erlangung des akademischen Grades Doktor-Ingenieur (Dr.-Ing.) vorgelegte Dissertation von Dipl.-Ing. Jens Kober aus Künzelsau März2012—Darmstadt—D17 Learning Motor Skills: From Algorithms to Robot Experiments Erlernen Motorischer Fähigkeiten: Von Algorithmen zu Roboter-Experimenten Vorgelegte Dissertation von Dipl.-Ing. Jens Kober aus Künzelsau 1. Gutachten: Prof. Dr. Jan Peters 2. Gutachten: Prof. Dr. Stefan Schaal Tag der Einreichung: Darmstadt—D17 Bitte zitieren Sie dieses Dokument als: URN: urn:nbn:de:tuda-tuprints-12345 URL: http://tuprints.ulb.tu-darmstadt.de/12345 Dieses Dokument wird bereitgestellt von tuprints, E-Publishing-Service der TU Darmstadt http://tuprints.ulb.tu-darmstadt.de [email protected] Die Veröffentlichung steht unter folgender Creative Commons Lizenz: Namensnennung – Keine kommerzielle Nutzung – Keine Bearbeitung 2.0 Deutschland http://creativecommons.org/licenses/by-nc-nd/2.0/de/ Erklärung zur Dissertation Hiermit versichere ich, die vorliegende Dissertation ohne Hilfe Dritter nur mit den an- gegebenen Quellen und Hilfsmitteln angefertigt zu haben. Alle Stellen, die aus Quellen entnommen wurden, sind als solche kenntlich gemacht. Diese Arbeit hat in gleicher oder ähnlicher Form noch keiner Prüfungsbehörde vorgelegen. Darmstadt, den 13. März 2012 (J. Kober) i Abstract Ever since the word “robot” was introduced to the English language by Karel Capek’sˇ play “Rossum’s Universal Robots” in 1921, robots have been expected to become part of our daily lives. In recent years, robots such as autonomous vacuum cleaners, lawn mowers, and window cleaners, as well as a huge number of toys have been made commercially available.
    [Show full text]
  • Exploring Applications of Deep Reinforcement Learning for Real-World Autonomous Driving Systems
    Exploring Applications of Deep Reinforcement Learning for Real-world Autonomous Driving Systems Victor Talpaert1;2, Ibrahim Sobh3, B. Ravi Kiran2, Patrick Mannion4, Senthil Yogamani5, Ahmad El-Sallab3 and Patrick Perez6 1U2IS, ENSTA ParisTech, Palaiseau, France 2AKKA Technologies, Guyancourt, France 3Valeo Egypt, Cairo 4Galway-Mayo Institute of Technology, Ireland 5Valeo Vision Systems, Ireland 6Valeo.ai, France Keywords: Autonomous Driving, Deep Reinforcement Learning, Visual Perception. Abstract: Deep Reinforcement Learning (DRL) has become increasingly powerful in recent years, with notable achie- vements such as Deepmind’s AlphaGo. It has been successfully deployed in commercial vehicles like Mo- bileye’s path planning system. However, a vast majority of work on DRL is focused on toy examples in controlled synthetic car simulator environments such as TORCS and CARLA. In general, DRL is still at its infancy in terms of usability in real-world applications. Our goal in this paper is to encourage real-world deployment of DRL in various autonomous driving (AD) applications. We first provide an overview of the tasks in autonomous driving systems, reinforcement learning algorithms and applications of DRL to AD sys- tems. We then discuss the challenges which must be addressed to enable further progress towards real-world deployment. 1 INTRODUCTION cific control task. The rest of the paper is structured as follows. Section 2 provides background on the vari- Autonomous driving (AD) is a challenging applica- ous modules of AD, an overview of reinforcement le- tion domain for machine learning (ML). Since the arning and a summary of applications of DRL to AD. task of driving “well” is already open to subjective Section 3 discusses challenges and open problems in definitions, it is not easy to specify the correct be- applying DRL to AD.
    [Show full text]
  • One-Shot Imitation Learning
    One-Shot Imitation Learning Yan Duanyx, Marcin Andrychowiczz, Bradly Stadieyz, Jonathan Hoyx, Jonas Schneiderz, Ilya Sutskeverz, Pieter Abbeelyx, Wojciech Zarembaz yBerkeley AI Research Lab, zOpenAI xWork done while at OpenAI {rockyduan, jonathanho, pabbeel}@eecs.berkeley.edu {marcin, bstadie, jonas, ilyasu, woj}@openai.com Abstract Imitation learning has been commonly applied to solve different tasks in isolation. This usually requires either careful feature engineering, or a significant number of samples. This is far from what we desire: ideally, robots should be able to learn from very few demonstrations of any given task, and instantly generalize to new situations of the same task, without requiring task-specific engineering. In this paper, we propose a meta-learning framework for achieving such capability, which we call one-shot imitation learning. Specifically, we consider the setting where there is a very large (maybe infinite) set of tasks, and each task has many instantiations. For example, a task could be to stack all blocks on a table into a single tower, another task could be to place all blocks on a table into two-block towers, etc. In each case, different instances of the task would consist of different sets of blocks with different initial states. At training time, our algorithm is presented with pairs of demonstrations for a subset of all tasks. A neural net is trained such that when it takes as input the first demonstration demonstration and a state sampled from the second demonstration, it should predict the action corresponding to the sampled state. At test time, a full demonstration of a single instance of a new task is presented, and the neural net is expected to perform well on new instances of this new task.
    [Show full text]
  • Reward Learning from Human Preferences and Demonstrations in Atari
    Reward learning from human preferences and demonstrations in Atari Borja Ibarz Jan Leike Tobias Pohlen DeepMind DeepMind DeepMind [email protected] [email protected] [email protected] Geoffrey Irving Shane Legg Dario Amodei OpenAI DeepMind OpenAI [email protected] [email protected] [email protected] Abstract To solve complex real-world problems with reinforcement learning, we cannot rely on manually specified reward functions. Instead, we can have humans communicate an objective to the agent directly. In this work, we combine two approaches to learning from human feedback: expert demonstrations and trajectory preferences. We train a deep neural network to model the reward function and use its predicted reward to train an DQN-based deep reinforcement learning agent on 9 Atari games. Our approach beats the imitation learning baseline in 7 games and achieves strictly superhuman performance on 2 games without using game rewards. Additionally, we investigate the goodness of fit of the reward model, present some reward hacking problems, and study the effects of noise in the human labels. 1 Introduction Reinforcement learning (RL) has recently been very successful in solving hard problems in domains with well-specified reward functions (Mnih et al., 2015, 2016; Silver et al., 2016). However, many tasks of interest involve goals that are poorly defined or hard to specify as a hard-coded reward. In those cases we can rely on demonstrations from human experts (inverse reinforcement learning, Ng and Russell, 2000; Ziebart et al., 2008), policy feedback (Knox and Stone, 2009; Warnell et al., 2017), or trajectory preferences (Wilson et al., 2012; Christiano et al., 2017).
    [Show full text]
  • Learning Robust Rewards with Adversarial Inverse Reinforcement Learning
    Published as a conference paper at ICLR 2018 LEARNING ROBUST REWARDS WITH ADVERSARIAL INVERSE REINFORCEMENT LEARNING Justin Fu, Katie Luo, Sergey Levine Department of Electrical Engineering and Computer Science University of California, Berkeley Berkeley, CA 94720, USA [email protected],[email protected], [email protected] ABSTRACT Reinforcement learning provides a powerful and general framework for decision making and control, but its application in practice is often hindered by the need for extensive feature and reward engineering. Deep reinforcement learning meth- ods can remove the need for explicit engineering of policy or value features, but still require a manually specified reward function. Inverse reinforcement learning holds the promise of automatic reward acquisition, but has proven exceptionally difficult to apply to large, high-dimensional problems with unknown dynamics. In this work, we propose AIRL, a practical and scalable inverse reinforcement learn- ing algorithm based on an adversarial reward learning formulation. We demon- strate that AIRL is able to recover reward functions that are robust to changes in dynamics, enabling us to learn policies even under significant variation in the environment seen during training. Our experiments show that AIRL greatly out- performs prior methods in these transfer settings. 1 INTRODUCTION While reinforcement learning (RL) provides a powerful framework for automating decision making and control, significant engineering of elements such as features and reward functions has typically been required for good practical performance. In recent years, deep reinforcement learning has al- leviated the need for feature engineering for policies and value functions, and has shown promising results on a range of complex tasks, from vision-based robotic control (Levine et al., 2016) to video games such as Atari (Mnih et al., 2015) and Minecraft (Oh et al., 2016).
    [Show full text]
  • Aerobatics Control of Flying Creatures Via Self-Regulated Learning
    Aerobatics Control of Flying Creatures via Self-Regulated Learning JUNGDAM WON, Seoul National University, South Korea JUNGNAM PARK, Seoul National University, South Korea JEHEE LEE∗, Seoul National University, South Korea Fig. 1. The imaginary dragon learned to perform aerobatic maneuvers. The dragon is physically simulated in realtime and interactively controllable. Flying creatures in animated films often perform highly dynamic aerobatic 1 INTRODUCTION maneuvers, which require their extreme of exercise capacity and skillful In animated films, flying creatures such as birds and dragons often control. Designing physics-based controllers (a.k.a., control policies) for perform highly dynamic flight skills, called aerobatics. The skills aerobatic maneuvers is very challenging because dynamic states remain in unstable equilibrium most of the time during aerobatics. Recently, Deep Re- include rapid rotation about three major axes (pitch, yaw, roll), and inforcement Learning (DRL) has shown its potential in constructing physics- a sequence of skills are performed in a consecutive manner to make based controllers. In this paper, we present a new concept, Self-Regulated dramatic effects. The creature manages to complete aerobatic skills Learning (SRL), which is combined with DRL to address the aerobatics con- by using its extreme of exercise capacity and endurance, which trol problem. The key idea of SRL is to allow the agent to take control over make it attractive and create tension in the audience. its own learning using an additional self-regulation policy. The policy allows Designing a physics-based controller for aerobatics is very chal- the agent to regulate its goals according to the capability of the current con- lenging because it requires extremely skillful control.
    [Show full text]