Training robots with physical bodies requires developing new methods and action representations that allow the learning agents to explore the space of policies efficiently. This work studies sample-efficient learning of complex policies in the context of robot table tennis. It incorporates learning into a hierarchical control framework using a model-free strategy layer (which requires complex reasoning about opponents that is difficult to do in a model-based way), model-based prediction of external objects (which are difficult to control directly with analytic control methods, but governed by learnable and relatively simple laws of physics), and analytic controllers for the robot itself. Human demonstrations are used to train dynamics models, which together with the analytic controller allow any robot that is physically capable to play table tennis without training episodes. Using only about 7,000 demonstrated trajectories, a striking policy can hit ball targets with about 20 cm error. Self-play is used to train cooperative and adversarial strategies on top of model-based striking skills trained from human demonstrations. After only about 24,000 strikes in self-play the agent learns to best exploit the human dynamics models for longer cooperative games. Further experiments demonstrate that more flexible variants of the policy can discover new strikes not demonstrated by humans and achieve higher performance at the expense of lower sample-efficiency. Experiments are carried out in a virtual reality environment using sensory observations that are obtainable in the real world. The high sample-efficiency demonstrated in the evaluations show that the proposed method is suitable for learning directly on physical robots without transfer of models or policies from simulation. Supplementary material available at https://sites.google.com/view/robottabletennis
Deep reinforcement learning provides a promising approach for vision-based control of real-world robots. However, the generalization of such models depends critically on the quantity and variety of data available for training. This data can be difficult to obtain for some types of robotic systems, such as fragile, small-scale quadrotors. Simulated rendering and physics can provide for much larger datasets, but such data is inherently of lower quality: many of the phenomena that make the real-world autonomous flight problem challenging, such as complex physics and air currents, are modeled poorly or not at all, and the systematic differences between simulation and the real world are typically impossible to eliminate. In this work, we investigate how data from both simulation and the real world can be combined in a hybrid deep reinforcement learning algorithm. Our method uses real-world data to learn about the dynamics of the system, and simulated data to learn a generalizable perception system that can enable the robot to avoid collisions using only a monocular camera. We demonstrate our approach on a real-world nano aerial vehicle collision avoidance task, showing that with only an hour of real-world data, the quadrotor can avoid collisions in new environments with various lighting conditions and geometry. Code, instructions for building the aerial vehicles, and videos of the experiments can be found at github.com/gkahn13/GtS
A central challenge in reinforcement learning is discovering effective policies for tasks where rewards are sparsely distributed. We postulate that in the absence of useful reward signals, an effective exploration strategy should seek out {\it decision states}. These states lie at critical junctions in the state space from where the agent can transition to new, potentially unexplored regions. We propose to learn about decision states from prior experience. By training a goal-conditioned policy with an information bottleneck, we can identify decision states by examining where the model actually leverages the goal state. We find that this simple mechanism effectively identifies decision states, even in partially observed settings. In effect, the model learns the sensory cues that correlate with potential subgoals. In new environments, this model can then identify novel subgoals for further exploration, guiding the agent through a sequence of potential decision states and through new regions of the state space.
We introduce a neural architecture for navigation in novel environments. Our proposed architecture learns to map from first-person views and plans a sequence of actions towards goals in the environment. The Cognitive Mapper and Planner (CMP) is based on two key ideas: a) a unified joint architecture for mapping and planning, such that the mapping is driven by the needs of the task, and b) a spatial memory with the ability to plan given an incomplete set of observations about the world. CMP constructs a top-down belief map of the world and applies a differentiable neural net planner to produce the next action at each time step. The accumulated belief of the world enables the agent to track visited regions of the environment. We train and test CMP on navigation problems in simulation environments derived from scans of real world buildings. Our experiments demonstrate that CMP outperforms alternate learning-based architectures, as well as, classical mapping and path planning approaches in many cases. Furthermore, it naturally extends to semantically specified goals, such as 'going to a chair'. We also deploy CMP on physical robots in indoor environments, where it achieves reasonable performance, even though it is trained entirely in simulation.
In the NeurIPS 2018 Artificial Intelligence for Prosthetics challenge, participants were tasked with building a controller for a musculoskeletal model with a goal of matching a given time-varying velocity vector. Top participants were invited to describe their algorithms. In this work, we describe the challenge and present thirteen solutions that used deep reinforcement learning approaches. Many solutions use similar relaxations and heuristics, such as reward shaping, frame skipping, discretization of the action space, symmetry, and policy blending. However, each team implemented different modifications of the known algorithms by, for example, dividing the task into subtasks, learning low-level control, or by incorporating expert knowledge and using imitation learning.
Imitation learning provides an appealing framework for autonomous control: in many tasks, demonstrations of preferred behavior can be readily obtained from human experts, removing the need for costly and potentially dangerous online data collection in the real world. However, policies learned with imitation learning have limited flexibility to accommodate varied goals at test-time. Model-based reinforcement learning (MBRL) offers considerably more flexibility, since a predictive model learned from data can be used to achieve various goals at test-time. However, MBRL suffers from two shortcomings. First, the model does not help to choose desired or safe outcomes -- its dynamics estimate only what is possible, not what is preferred. Second, MBRL typically requires additional online data collection to ensure that the model is accurate in those situations that are actually encountered when attempting to achieve test-time goals. Collecting this data with a partially trained model can be dangerous and time-consuming. In this paper, we aim to combine the benefits of imitation learning and MBRL, and propose imitative models: probabilistic predictive models able to plan expert-like trajectories to achieve arbitrary goals. We find this method substantially outperforms both direct imitation and MBRL in a simulated autonomous driving task, and can be learned efficiently from a fixed set of expert demonstrations without additional online data collection. We also show our model can flexibly incorporate user-supplied costs at test-time, can plan to sequences of goals, and can even perform well with imprecise goals, including goals on the wrong side of the road.
Model-free deep reinforcement learning (RL) algorithms have been successfully applied to a range of challenging sequential decision making and control tasks. However, these methods typically suffer from two major challenges: high sample complexity and brittleness to hyperparameters. Both of these challenges limit the applicability of such methods to real-world domains. In this paper, we describe Soft Actor-Critic (SAC), our recently introduced off-policy actor-critic algorithm based on the maximum entropy RL framework. In this framework, the actor aims to simultaneously maximize expected return and entropy. That is, to succeed at the task while acting as randomly as possible. We extend SAC to incorporate a number of modifications that accelerate training and improve stability with respect to the hyperparameters, including a constrained formulation that automatically tunes the temperature hyperparameter. We systematically evaluate SAC on a range of benchmark tasks, as well as real-world challenging tasks such as locomotion for a quadrupedal robot and robotic manipulation with a dexterous hand. With these improvements, SAC achieves state-of-the-art performance, outperforming prior on-policy and off-policy methods in sample-efficiency and asymptotic performance. Furthermore, we demonstrate that, in contrast to other off-policy algorithms, our approach is very stable, achieving similar performance across different random seeds. These results suggest that SAC is a promising candidate for learning in real-world robotics tasks.
Humans and animals can learn complex predictive models that allow them to accurately and reliably reason about real-world phenomena, and they can adapt such models extremely quickly in the face of unexpected changes. Deep neural network models allow us to represent very complex functions, but lack this capacity for rapid online adaptation. The goal in this paper is to develop a method for continual online learning from an incoming stream of data, using deep neural network models. We formulate an online learning procedure that uses stochastic gradient descent to update model parameters, and an expectation maximization algorithm with a Chinese restaurant process prior to develop and maintain a mixture of models to handle non-stationary task distributions. This allows for all models to be adapted as necessary, with new models instantiated for task changes and old models recalled when previously seen tasks are encountered again. Furthermore, we observe that meta-learning can be used to meta-train a model such that this direct online adaptation with SGD is effective, which is otherwise not the case for large function approximators. In this work, we apply our meta-learning for online learning (MOLe) approach to model-based reinforcement learning, where adapting the predictive model is critical for control; we demonstrate that MOLe outperforms alternative prior methods, and enables effective continuous adaptation in non-stationary task distributions such as varying terrains, motor failures, and unexpected disturbances.
Generating low-level robot controllers often requires manual parameters tuning and significant system knowledge, which can result in long design times for highly specialized controllers. With the growth of automation, the need for such controllers might grow faster than the number of expert designers. To address the problem of rapidly generating low-level controllers without domain knowledge, we propose using model-based reinforcement learning (MBRL) trained on few minutes of automatically generated data. In this paper, we explore the capabilities of MBRL on a Crazyflie quadrotor with rapid dynamics where existing classical control schemes offer a baseline against the new method's performance. To our knowledge, this is the first use of MBRL for low-level controlled hover of a quadrotor using only on-board sensors, direct motor input signals, and no initial dynamics knowledge. Our forward dynamics model for prediction is a neural network tuned to predict the state variables at the next time step, with a regularization term on the variance of predictions. The model predictive controller then transmits best actions from a GPU-enabled base station to the quadrotor firmware via radio. In our experiments, the quadrotor achieved hovering capability of up to 6 seconds with 3 minutes of experimental training data.
Object-based factorizations provide a useful level of abstraction for interacting with the world. Building explicit object representations, however, often requires supervisory signals that are difficult to obtain in practice. We present a paradigm for learning object-centric representations for physical scene understanding without direct supervision of object properties. Our model, Object-Oriented Prediction and Planning (O2P2), jointly learns a perception function to map from image observations to object representations, a pairwise physics interaction function to predict the time evolution of a collection of objects, and a rendering function to map objects back to pixels. For evaluation, we consider not only the accuracy of the physical predictions of the model, but also its utility for downstream tasks that require an actionable representation of intuitive physics. After training our model on an image prediction task, we can use its learned representations to build block towers more complicated than those observed during training.