An optimal feedback controller for a given Markov decision process (MDP) can in principle be synthesized by value or policy iteration. However, if the system dynamics and the reward function are unknown, a learning agent must discover an optimal controller via direct interaction with the environment. Such interactive data gathering commonly leads to divergence towards dangerous or uninformative regions of the state space unless additional regularization measures are taken. Prior works proposed bounding the information loss measured by the Kullback-Leibler (KL) divergence at every policy improvement step to eliminate instability in the learning dynamics. In this paper, we consider a broader family of $f$-divergences, and more concretely $\alpha$-divergences, which inherit the beneficial property of providing the policy improvement step in closed form at the same time yielding a corresponding dual objective for policy evaluation. Such entropic proximal policy optimization view gives a unified perspective on compatible actor-critic architectures. In particular, common least-squares value function estimation coupled with advantage-weighted maximum likelihood policy improvement is shown to correspond to the Pearson $\chi^2$-divergence penalty. Other actor-critic pairs arise for various choices of the penalty-generating function $f$. On a concrete instantiation of our framework with the $\alpha$-divergence, we carry out asymptotic analysis of the solutions for different values of $\alpha$ and demonstrate the effects of the divergence function choice on common standard reinforcement learning problems.
Learning robot control policies from physics simulations is of great interest to the robotics community as it may render the learning process faster, cheaper, and safer by alleviating the need for expensive real-world experiments. However, the direct transfer of learned behavior from simulation to reality is a major challenge. Optimizing a policy on a slightly faulty simulator can easily lead to the maximization of the 'Simulation Optimization Bias' (SOB). In this case, the optimizer exploits modeling errors of the simulator such that the resulting behavior can potentially damage the robot. We tackle this challenge by applying domain randomization, i.e., randomizing the parameters of the physics simulations during learning. We propose an algorithm called Simulation-based Policy Optimization with Transferability Assessment (SPOTA) which uses an estimator of the SOB to formulate a stopping criterion for training. The introduced estimator quantifies the over-fitting to the set of domains experienced while training. Our experimental results in two different environments show that the new simulation-based policy search algorithm is able to learn a control policy exclusively from a randomized simulator, which can be applied directly to real system without any additional training on the latter.
Deep learning has achieved astonishing results on many tasks with large amounts of data and generalization within the proximity of training data. For many important real-world applications, these requirements are unfeasible and additional prior knowledge on the task domain is required to overcome the resulting problems. In particular, learning physics models for model-based control requires robust extrapolation from fewer samples - often collected online in real-time - and model errors may lead to drastic damages of the system. Directly incorporating physical insight has enabled us to obtain a novel deep model learning approach that extrapolates well while requiring fewer samples. As a first example, we propose Deep Lagrangian Networks (DeLaN) as a deep network structure upon which Lagrangian Mechanics have been imposed. DeLaN can learn the equations of motion of a mechanical system (i.e., system dynamics) with a deep network efficiently while ensuring physical plausibility. The resulting DeLaN network performs very well at robot tracking control. The proposed method did not only outperform previous model learning approaches at learning speed but exhibits substantially improved and more robust extrapolation to novel trajectories and learns online in real-time
Applying Deep Learning to control has a lot of potential for enabling the intelligent design of robot control laws. Unfortunately common deep learning approaches to control, such as deep reinforcement learning, require an unrealistic amount of interaction with the real system, do not yield any performance guarantees, and do not make good use of extensive insights from model-based control. In particular, common black-box approaches -- that abandon all insight from control -- are not suitable for complex robot systems. We propose a deep control approach as a bridge between the solid theoretical foundations of energy-based control and the flexibility of deep learning. To accomplish this goal, we extend Deep Lagrangian Networks (DeLaN) to not only adhere to Lagrangian Mechanics but also ensure conservation of energy and passivity of the learned representation. This novel extension is embedded within generic model-based control laws to enable energy control of under-actuated systems. The resulting DeLaN for energy control (DeLaN 4EC) is the first model learning approach using generic function approximation that is capable of learning energy control. DeLaN 4EC exhibits excellent real-time control on the physical Furuta Pendulum and learns to swing-up the pendulum while the control law using system identification does not.
Similarity distance measure between two trajectories is an essential tool to understand patterns in motion, for example, in Human-Robot Interaction or Imitation Learning. The problem has been faced in many fields, from Signal Processing, Probabilistic Theory field, Topology field or Statistics field.Anyway, up to now, none of the trajectory similarity measurements metrics are invariant to all possible linear transformation of the trajectories (rotation, scaling, reflection, shear mapping or squeeze mapping). Also not all of them are robust in front of noisy signals or fast enough for real-time trajectory classification. To overcome this limitation this paper proposes a similarity distance metric that will remain invariant in front of any possible linear transformation.Based on Pearson Correlation Coefficient and the Coefficient of Determination, our similarity metric, the Generalized Multiple Correlation Coefficient (GMCC) is presented like the natural extension of the Multiple Correlation Coefficient. The motivation of this paper is two fold. First, to introduce a new correlation metric that presents the best properties to compute similarities between trajectories invariant to linear transformations and compare it with some state of the art similarity distances.Second, to present a natural way of integrating the similarity metric in an Imitation Learning scenario for clustering robot trajectories.
Assistive robots can potentially improve the quality of life and personal independence of elderly people by supporting everyday life activities. To guarantee a safe and intuitive interaction between human and robot, human intentions need to be recognized automatically. As humans communicate their intentions multimodally, the use of multiple modalities for intention recognition may not just increase the robustness against failure of individual modalities but especially reduce the uncertainty about the intention to be predicted. This is desirable as particularly in direct interaction between robots and potentially vulnerable humans a minimal uncertainty about the situation as well as knowledge about this actual uncertainty is necessary. Thus, in contrast to existing methods, in this work a new approach for multimodal intention recognition is introduced that focuses on uncertainty reduction through classifier fusion. For the four considered modalities speech, gestures, gaze directions and scene objects individual intention classifiers are trained, all of which output a probability distribution over all possible intentions. By combining these output distributions using the Bayesian method Independent Opinion Pool the uncertainty about the intention to be recognized can be decreased. The approach is evaluated in a collaborative human-robot interaction task with a 7-DoF robot arm. The results show that fused classifiers which combine multiple modalities outperform the respective individual base classifiers with respect to increased accuracy, robustness, and reduced uncertainty.
Similarity distance measure between two trajectories is an essential tool to understand patterns in motion, for example, in Human-Robot Interaction or Imitation Learning. The problem has been faced in many fields, from Signal Processing, Probabilistic Theory field, Topology field or Statistics field.Anyway, up to now, none of the trajectory similarity measurements metrics are invariant to all possible linear transformation of the trajectories (rotation, scaling, reflection, shear mapping or squeeze mapping). Also not all of them are robust in front of noisy signals or fast enough for real-time trajectory classification. To overcome this limitation this paper proposes a similarity distance metric that will remain invariant in front of any possible linear transformation.Based on Pearson Correlation Coefficient and the Coefficient of Determination, our similarity metric, the Generalized Multiple Correlation Coefficient (GMCC) is presented like the natural extension of the Multiple Correlation Coefficient. The motivation of this paper is two fold. First, to introduce a new correlation metric that presents the best properties to compute similarities between trajectories invariant to linear transformations and compare it with some state of the art similarity distances.Second, to present a natural way of integrating the similarity metric in an Imitation Learning scenario for clustering robot trajectories.
With the increasing pace of automation, modern robotic systems need to act in stochastic, non-stationary, partially observable environments. A range of algorithms for finding parameterized policies that optimize for long-term average performance have been proposed in the past. However, the majority of the proposed approaches does not explicitly take into account the variability of the performance metric, which may lead to finding policies that although performing well on average, can perform spectacularly bad in a particular run or over a period of time. To address this shortcoming, we study an approach to policy optimization that explicitly takes into account higher order statistics of the reward function. In this paper, we extend policy gradient methods to include the entropic risk measure in the objective function and evaluate their performance in simulation experiments and on a real-robot task of learning a hitting motion in robot badminton.
Motor babbling and goal babbling has been used for sensorimotor learning of highly redundant systems in soft robotics. Recent works in goal babbling has demonstrated successful learning of inverse kinematics (IK) on such systems, and suggests that babbling in the goal space better resolves motor redundancy by learning as few sensorimotor mapping as possible. However, for musculoskeletal robot systems, motor redundancy can be of useful information to explain muscle activation patterns, thus the term motor abundance. In this work, we introduce some simple heuristics to empirically define the unknown goal space, and learn the inverse kinematics of a 10 DoF musculoskeletal robot arm using directed goal babbling. We then further propose local online motor babbling using Covariance Matrix Adaptation Evolution Strategy (CMA-ES), which bootstraps on the collected samples in goal babbling for initialization, such that motor abundance can be queried for any static goal within the defined goal space. The result shows that our motor babbling approach can efficiently explore motor abundance, and gives useful insights in terms of muscle stiffness and synergy.