Learning tasks from a single demonstration

By Christopher G. Atkeson, Stefan Schaal. 1997

Subfields: Reinforcement Learning, Robotics & Control

How researchers see this paper

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

In Reinforcement Learning

In Robotics & Control

Citations in the corpus

When to Trust Your Model: Model-Based Policy Optimization (2019, methodology): “The bottleneck in scaling model-based approaches to complex tasks often lies in learning reliable predictive models of high-dimensional dynamics (Atkeson & Schaal, 1997).”
Emotion and sociable humanoid robots (2003, background): “Learning by demonstration to acquire physical skills such as pole balancing (Atkeson and Schaal, 1997a,b; Schaal, 1997), learning by imitation to acquire a proto-language (Billard, 2002), and learning to imitate in order to produce a sequence of gestures (Demiris and Hayes, 2002; Mataric, 2000)…”
Policy Shaping: Integrating Human Feedback with Reinforcement Learning (2013, background): “Several papers have shown how the policy information in human demonstrations can be used for inverse optimal control [14, 15], to seed an agent’s exploration [16, 17], and in some cases be used entirely in place of exploration [18, 19].”
Learning to Walk in 20 Minutes (2005, methodology): “Atkeson and Schaal used learning from demonstration to train an anthropomorphic robotic arm to accomplish a pendulum swing-up task[1].”
Learning from observation using primitives (2001, background): “such as trajectory learning [1], learning from observation [4, 5, 11, 15, 6, 10, 12, 13], postural primitives [21], and other techniques that decompose complex tasks or movements into smaller parts [2, 7, 16].”
Robot Learning From Demonstration (1997, unknown): “Atkeson and Schaal 1997 report on a preliminary implementation of learning from demonstration for the pendulum swing up task.”
Exploration from Demonstration for Interactive Reinforcement Learning (2016, background): “Several papers have shown how the policy information in human demonstrations can be used for inverse optimal control [28, 2], for teaching [8], to seed an agent’s exploration [5, 38], and in some cases be used entirely in place of exploration [21, 33].”
Data-Efficient Learning of Feedback Policies from Image Pixels using Deep Dynamical Models (2015, methodology): “This data inefﬁciency is not speciﬁc to Q-learning, but a general problem of model-free RL methods [3, 14].”
Discovering optimal imitation strategies (2004, background): “…hand–object actions (picking up a block, rotating the block) [9–11,17], state-actions (turn left, move forward) [6] [7], the path followed by the manipulated object [2] [14] and the joint trajectories of the demonstrator's motion [4] [5] [8], which could be reproduced by pre-defined motor programs.”
A comparison of direct and model-based reinforcement learning (1997, methodology): “Atkeson and Schaal (1997) describe an implementation of model-based reinforcement learning of a pendulum swing up on an actual robot that identies a parametric pendulum model.”
Computational approaches to motor learning by imitation. (2003, unknown): “The second matching paradigm is based on the idea of predictive forward models (Miall & Wolpert 1996; Schaal 1997; Atkeson & Schaal 1997a; Wolpert et al. 1998, 2003; Demiris & Hayes 2002).”
Planning and Execution using Inaccurate Models with Provable Guarantees (2020, methodology): “Using local function approximation methods in robotics has been heavily explored in seminal works [31, 2] due to their smaller sample complexity requirements and local generalization properties that do not cause interference [3, 8].”
Full Paper Reinforcement Learning for Imitating Constrained Reaching Movements (unknown): “RL algorithms have already been successfully implemented in robotic applications such as swinging up and controlling an inverse pendulum [7], refining forehand and backhand tennis swing [8] or acquiring a dynamic stand-up behavior [9, 10].”