Reinforcement Learning for Humanoid Robotics

By Jan Peters, Sethu Vijayakumar, Stefan Schaal. 2003

404 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Active learning of inverse models with intrinsically motivated goal exploration in robots (2013, positive): “For instance,studiespresentedin[13]combineRLwiththeframeworkof learning by demonstration.”
Batch Mode Reinforcement Learning based on the Synthesis of Artificial Trajectories (2012, positive): “Applications of these batch mode RL techniques have already led to promising results in robotics (Peters et al. 2003; Bonarini et al. 2008; Tognetti et al. 2009), power systems (Ernst et al. 2009), image processing (Ernst et al. 2006a), water reservoir optimization (Castelletti et al. 2007, 2010),…”
Intrinsically motivated goal exploration for active motor learning in robots: A case study (2010, unknown): “For instance, studies presented by Peters et al. [ 5 ] present a framework that combines learning by demonstration with reinforcement learning techniques: the engineer defines a specific goal in the task space and a corresponding handcrafted reward function, and a human demonstrator provides an exam”
Reinforcement learning of motor skills with policy gradients (2008, methodology): “We summarize our results from Peters et al. (2003) and outline the derivation of Fisher information of paths here.”
Bayesian Reinforcement Learning (2010, background): “In empirical evaluations, natural PG has been shown to signiﬁcantly outperform conventional PG (Kakade, 2002; Bagnell and Schneider, 2003; Peters et al, 2003; Peters and Schaal, 2008).”
Approximate Inference and Stochastic Optimal Control (2010, methodology): “, [6, 20], and we emphasise that, as the state dependent part of the basis is largely unrestricted, this general class of basis sets does not seem unreasonably restrictive.”
Actor-critic reinforcement learning for tracking control in robotics (2016, background): “Several well-known examples include autonomous helicopter control [6], humanoid robots [7], and manipulators [8].”
Incremental Natural Actor-Critic Algorithms (2007, methodology): “Another approach to speeding up policy gradient algorithms was proposed by Kakade (2002) and then refined and extended by Bagnell and Schneider (2003) and by Peters et al. (2003).”
Online Model Learning Algorithms for Actor-Critic Control (2015, methodology): “Inevitably, this resulted in work on actor-critic methods using an LSTD critic (Peters et al., 2003; Williams et al., 2006).”
Learning CPG Sensory Feedback with Policy Gradient for Biped Locomotion for a Full-Body Humanoid (2005, methodology): “The variance of the policy is modelled by a sigmoidal function (Kimura & Kobayashi 1998; Peters, Vijayakumar, & Schaal 2003).”
Algorithms for Reinforcement Learning (2010, methodology): “Hence, setting Vt+1 = Vθ(Yt+1) does not introduce any bias, while it is expected to reduce variance since Vt+1 does not depend on the randomness of A ′ t+1 (Peters et al., 2003; Peters and Schaal, 2008).16 We will further discuss this choice in the next section.”
Robot learning from demonstration by constructing skill trees (2012, methodology): “These methods have found wide applicability in robot applications (Ng et al. 2003; Peters et al. 2003; Kohl and Stone 2004; Tedrake et al. 2004; Peters and Schaal 2008; Neumann et al. 2009; Kober and Peters 2010) because they can easily represent policies defined over continuous actions and can…”
Regularization in reinforcement learning (2011, methodology): “Moreover, there are hybrid methods that explicitly represent both value and policy functions [Konda and Tsitsiklis, 2001; Peters et al., 2003; Ghavamzadeh and Engel, 2007a].”
Ieee Transactions on Systems, Man and Cybernetics—part C: Applications and Reviews 1 a Survey of Actor-critic Reinforcement Learning: Standard and Natural Policy Gradients (unknown): “In [52], it is noted that, in light of the policy gradient theorem that was only published many years later, Gullapalli's earlier idea in [32] of using the temporal difference δ in the gradient used to update the policy weights can be shown to yield the true policy gradient ∇ ϑ J(ϑ), and hence…”
A Survey on Policy Search for Robotics (2013, methodology): “This insight is used by the natural policy gradient algorithms [62, 63, 79] and by the REPS algorithm [17, 58].”