skip to main content

Title: Learning Task-Driven Control Policies via Information Bottlenecks
This paper presents a reinforcement learning approach to synthesizing task-driven control policies for robotic systems equipped with rich sensory modalities (e.g., vision or depth). Standard reinforcement learning algorithms typically produce policies that tightly couple control actions to the entirety of the system's state and rich sensor observations. As a consequence, the resulting policies can often be sensitive to changes in task-irrelevant portions of the state or observations (e.g., changing background colors). In contrast, the approach we present here learns to create a task-driven representation that is used to compute control actions. Formally, this is achieved by deriving a policy gradient-style algorithm that creates an information bottleneck between the states and the task-driven representation; this constrains actions to only depend on task-relevant information. We demonstrate our approach in a thorough set of simulation results on multiple examples including a grasping task that utilizes depth images and a ball-catching task that utilizes RGB images. Comparisons with a standard policy gradient approach demonstrate that the task-driven policies produced by our algorithm are often significantly more robust to sensor noise and task-irrelevant changes in the environment.
Award ID(s):
Publication Date:
Journal Name:
Robotics: Science and Systems (RSS)
Sponsoring Org:
National Science Foundation
More Like this
  1. Our goal is to develop a principled and general algorithmic framework for task-driven estimation and control for robotic systems. State-of-the-art approaches for controlling robotic systems typically rely heavily on accurately estimating the full state of the robot (e.g., a running robot might estimate joint angles and velocities, torso state, and position relative to a goal). However, full state representations are often excessively rich for the specific task at hand and can lead to significant computational inefficiency and brittleness to errors in state estimation. In contrast, we present an approach that eschews such rich representations and seeks to create task-driven representations.more »The key technical insight is to leverage the theory of information bottlenecks}to formalize the notion of a "task-driven representation" in terms of information theoretic quantities that measure the minimality of a representation. We propose novel iterative algorithms for automatically synthesizing (offline) a task-driven representation (given in terms of a set of task-relevant variables (TRVs)) and a performant control policy that is a function of the TRVs. We present online algorithms for estimating the TRVs in order to apply the control policy. We demonstrate that our approach results in significant robustness to unmodeled measurement uncertainty both theoretically and via thorough simulation experiments including a spring-loaded inverted pendulum running to a goal location.« less
  2. Our goal is to learn control policies for robots that provably generalize well to novel environments given a dataset of example environments. The key technical idea behind our approach is to leverage tools from generalization theory in machine learning by exploiting a precise analogy (which we present in the form of a reduction) between generalization of control policies to novel environments and generalization of hypotheses in the supervised learning setting. In particular, we utilize the probably approximately correct (PAC)-Bayes framework, which allows us to obtain upper bounds that hold with high probability on the expected cost of (stochastic) control policiesmore »across novel environments. We propose policy learning algorithms that explicitly seek to minimize this upper bound. The corresponding optimization problem can be solved using convex optimization (relative entropy programming in particular) in the setting where we are optimizing over a finite policy space. In the more general setting of continuously parameterized policies (e.g., neural network policies), we minimize this upper bound using stochastic gradient descent. We present simulated results of our approach applied to learning (1) reactive obstacle avoidance policies and (2) neural network-based grasping policies. We also present hardware results for the Parrot Swing drone navigating through different obstacle environments. Our examples demonstrate the potential of our approach to provide strong generalization guarantees for robotic systems with continuous state and action spaces, complicated (e.g., nonlinear) dynamics, rich sensory inputs (e.g., depth images), and neural network-based policies.

    « less
  3. Wallach, H (Ed.)
    We study the problem of programmatic reinforcement learning, in which policies are represented as short programs in a symbolic language. Programmatic policies can be more interpretable, generalizable, and amenable to formal verification than neural policies; however, designing rigorous learning approaches for such policies remains a challenge. Our approach to this challenge-a meta-algorithm called PROPEL-is based on three insights. First, we view our learning task as optimization in policy space, modulo the constraint that the desired policy has a programmatic representation, and solve this optimization problem using a form of mirror descent that takes a gradient step into the unconstrained policymore »space and then projects back onto the constrained space. Second, we view the unconstrained policy space as mixing neural and programmatic representations, which enables employing state-of-the-art deep policy gradient approaches. Third, we cast the projection step as program synthesis via imitation learning, and exploit contemporary combinatorial methods for this task. We present theoretical convergence results for PROPEL and empirically evaluate the approach in three continuous control domains. The experiments show that PROPEL can significantly outperform state-of-the-art approaches for learning programmatic policies.« less
  4. Artificial Neural Networks (ANNs) are currently being used as function approximators in many state-of-the-art Reinforcement Learning (RL) algorithms. Spiking Neural Networks (SNNs) have been shown to drastically reduce the energy consumption of ANNs by encoding information in sparse temporal binary spike streams, hence emulating the communication mechanism of biological neurons. Due to their low energy consumption, SNNs are considered to be important candidates as co-processors to be implemented in mobile devices. In this work, the use of SNNs as stochastic policies is explored under an energy-efficient first-to-spike action rule, whereby the action taken by the RL agent is determined bymore »the occurrence of the first spike among the output neurons. A policy gradient-based algorithm is derived considering a Generalized Linear Model (GLM) for spiking neurons. Experimental results demonstrate the capability of online trained SNNs as stochastic policies to gracefully trade energy consumption, as measured by the number of spikes, and control performance. Significant gains are shown as compared to the standard approach of converting an offline trained ANN into an SNN.« less
  5. This work incorporates the effects that hazardous environments have on sensing devices, in the guidance of mobile platforms with onboard sensors. Mobile sensors are utilized in the state reconstruction of spatiotemporally varying processes, often described by advection-diffusion PDEs. A typical sensor guidance policy is based on a gradient ascent scheme which repositions the sensors to spatial regions that have larger state estimation errors. If the cumulative measurements of the spatial process are used as a means to represent the effects of hazardous environments on the sensors, then the sensors are considered inoperable the instance the cumulative measurements exceed a device-specificmore »tolerance level. A binary guidance policy considered earlier repositioned the sensors to regions of larger values of the state estimation errors thus implementing an information-sensitive policy. The policy switched to an information-averse guidance the instance the cumulative effects exceeded a certain tolerance level. Such a binary policy switches the sensor velocity abruptly from a positive to a negative value. To alleviate these discontinuity effects, a ternary guidance policy is considered and which inserts a third guidance policy, the information-neutral policy, that smooths out the transitions from information-sensitive to information-averse guidance. A novelty in this ternary guidance has to do with the level-set approach which changes from a guidance towards large values of the state estimation error towards level sets of the state estimation error and eventually towards reduced values of the state estimation error. An example on an advection-diffusion PDE in 2D employing a single interior mobile sensor using both the binary and ternary guidance policies is used to demonstrate the effects of hazardous environments on both the sensor life expectancy and the performance of the state estimator.« less