We develop a hybrid control approach for robot learning based on combining learned predictive models with experience-based state-action policy mappings to improve the learning capabilities of robotic systems. Predictive models provide an understanding of the task and the physics (which improves sample-efficiency), while experience-based pol-icy mappings are treated as “muscle memory” that encode favorable actions as experiences that override planned actions. Hybrid control tools are used to create an algorithmic approach for combining learned predictive models with experience-based learning. Hybrid learning is presented as a method for efficiently learning motor skills by systematically combining and improving the performance of predictive models and experience-based policies. A deterministic variation of hybrid learning is derived and extended into a stochastic implementation that relaxes some of the key assumptions in the original derivation. Each variation is tested on experience-based learning methods (where the robot interacts with the environment to gain experience) as well as imitation learning methods(where experience is provided through demonstrations and tested in the environment). The results show that our method is capable of improving the performance and sample-efficiency of learning motor skills in a variety of experimental domains.
more »
« less
Physics-informed Dyna-style model-based deep reinforcement learning for dynamic control
Model-based reinforcement learning (MBRL) is believed to have much higher sample efficiency compared with model-free algorithms by learning a predictive model of the environment. However, the performance of MBRL highly relies on the quality of the learned model, which is usually built in a black-box manner and may have poor predictive accuracy outside of the data distribution. The deficiencies of the learned model may prevent the policy from being fully optimized. Although some uncertainty analysis-based remedies have been proposed to alleviate this issue, model bias still poses a great challenge for MBRL. In this work, we propose to leverage the prior knowledge of underlying physics of the environment, where the governing laws are (partially) known. In particular, we developed a physics-informed MBRL framework, where governing equations and physical constraints are used to inform the model learning and policy search. By incorporating the prior information of the environment, the quality of the learned model can be notably improved, while the required interactions with the environment are significantly reduced, leading to better sample efficiency and learning performance. The effectiveness and merit have been demonstrated over a handful of classic control problems, where the environments are governed by canonical ordinary/partial differential equations.
more »
« less
- NSF-PAR ID:
- 10311898
- Date Published:
- Journal Name:
- Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences
- Volume:
- 477
- Issue:
- 2255
- ISSN:
- 1364-5021
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
In offline reinforcement learning (RL), the goal is to learn a highly rewarding policy based solely on a dataset of historical interactions with the environment. The ability to train RL policies offline would greatly expand where RL can be applied, its data efficiency, and its experimental velocity. Prior work in offline RL has been confined almost exclusively to model-free RL approaches. In this work, we present MOReL, an algorithmic framework for model-based offline RL. This framework consists of two steps: (a) learning a pessimistic MDP (P-MDP) using the offline dataset; (b) learning a near-optimal policy in this P-MDP. The learned P-MDP has the property that for any policy, the performance in the real environment is approximately lower-bounded by the performance in the P-MDP. This enables it to serve as a good surrogate for purposes of policy evaluation and learning, and overcome common pitfalls of model-based RL like model exploitation. Theoretically, we show that MOReL is minimax optimal (up to log factors) for offline RL. Through experiments, we show that MOReL matches or exceeds state-of-the-art results in widely studied offline RL benchmarks. Moreover, the modular design of MOReL enables future advances in its components (e.g., in model learning, planning etc.) to directly translate into improvements for offline RL.more » « less
-
We develop an approach to improve the learning capabilities of robotic systems by combining learned predictive models with experience-based state-action policy mappings. Predictive models provide an understanding of the task and the dynamics, while experience-based (model-free) policy mappings encode favorable actions that override planned actions. We refer to our approach of systematically combining model-based and model-free learning methods as hybrid learning. Our approach efficiently learns motor skills and improves the performance of predictive models and experience-based policies. Moreover, our approach enables policies (both model-based and model-free) to be updated using any off-policy reinforcement learning method. We derive a deterministic method of hybrid learning by optimally switching between learning modalities. We adapt our method to a stochastic variation that relaxes some of the key assumptions in the original derivation. Our deterministic and stochastic variations are tested on a variety of robot control benchmark tasks in simulation as well as a hardware manipulation task. We extend our approach for use with imitation learning methods, where experience is provided through demonstrations, and we test the expanded capability with a real-world pick-and-place task. The results show that our method is capable of improving the performance and sample efficiency of learning motor skills in a variety of experimental domains.more » « less
-
Learning multi-agent system dynamics has been extensively studied for various real-world applications, such as molecular dynamics in biology, multi-body system in physics, and particle dynamics in material science. Most of the existing models are built to learn single system dynamics, which learn the dynamics from observed historical data and predict the future trajectory. In practice, however, we might observe multiple systems that are generated across different environments, which differ in latent exogenous factors such as temperature and gravity. One simple solution is to learn multiple environment-specific models, but it fails to exploit the potential commonalities among the dynamics across environments and offers poor prediction results where per-environment data is sparse or limited. Here, we present GG-ODE (Generalized Graph Ordinary Differential Equations), a machine learning framework for learning continuous multi-agent system dynamics across environments. Our model learns system dynamics using neural ordinary differential equations (ODE) parameterized by Graph Neural Networks (GNNs) to capture the continuous interaction among agents. We achieve the model generalization by assuming the dynamics across different environments are governed by common physics laws that can be captured via learning a shared ODE function. The distinct latent exogenous factors learned for each environment are incorporated into the ODE function to account for their differences. To improve model performance, we additionally design two regularization losses to (1) enforce the orthogonality between the learned initial states and exogenous factors via mutual information minimization; and (2) reduce the temporal variance of learned exogenous factors within the same system via contrastive learning. Experiments over various physical simulations show that our model can accurately predict system dynamics, especially in the long range, and can generalize well to new systems with few observations.more » « less
-
Green wireless networks Wake-up radio Energy harvesting Routing Markov decision process Reinforcement learning 1. Introduction With 14.2 billions of connected things in 2019, over 41.6 billions expected by 2025, and a total spending on endpoints and services that will reach well over $1.1 trillion by the end of 2026, the Internet of Things (IoT) is poised to have a transformative impact on the way we live and on the way we work [1–3]. The vision of this ‘‘connected continuum’’ of objects and people, however, comes with a wide variety of challenges, especially for those IoT networks whose devices rely on some forms of depletable energy support. This has prompted research on hardware and software solutions aimed at decreasing the depen- dence of devices from ‘‘pre-packaged’’ energy provision (e.g., batteries), leading to devices capable of harvesting energy from the environment, and to networks – often called green wireless networks – whose lifetime is virtually infinite. Despite the promising advances of energy harvesting technologies, IoT devices are still doomed to run out of energy due to their inherent constraints on resources such as storage, processing and communica- tion, whose energy requirements often exceed what harvesting can provide. The communication circuitry of prevailing radio technology, especially, consumes relevant amount of energy even when in idle state, i.e., even when no transmissions or receptions occur. Even duty cycling, namely, operating with the radio in low energy consumption ∗ Corresponding author. E-mail address: koutsandria@di.uniroma1.it (G. Koutsandria). https://doi.org/10.1016/j.comcom.2020.05.046 (sleep) mode for pre-set amounts of time, has been shown to only mildly alleviate the problem of making IoT devices durable [4]. An effective answer to eliminate all possible forms of energy consumption that are not directly related to communication (e.g., idle listening) is provided by ultra low power radio triggering techniques, also known as wake-up radios [5,6]. Wake-up radio-based networks allow devices to remain in sleep mode by turning off their main radio when no communication is taking place. Devices continuously listen for a trigger on their wake-up radio, namely, for a wake-up sequence, to activate their main radio and participate to communication tasks. Therefore, devices wake up and turn their main radio on only when data communication is requested by a neighboring device. Further energy savings can be obtained by restricting the number of neighboring devices that wake up when triggered. This is obtained by allowing devices to wake up only when they receive specific wake-up sequences, which correspond to particular protocol requirements, including distance from the destina- tion, current energy status, residual energy, etc. This form of selective awakenings is called semantic addressing [7]. Use of low-power wake-up radio with semantic addressing has been shown to remarkably reduce the dominating energy costs of communication and idle listening of traditional radio networking [7–12]. This paper contributes to the research on enabling green wireless networks for long lasting IoT applications. Specifically, we introduce a ABSTRACT This paper presents G-WHARP, for Green Wake-up and HARvesting-based energy-Predictive forwarding, a wake-up radio-based forwarding strategy for wireless networks equipped with energy harvesting capabilities (green wireless networks). Following a learning-based approach, G-WHARP blends energy harvesting and wake-up radio technology to maximize energy efficiency and obtain superior network performance. Nodes autonomously decide on their forwarding availability based on a Markov Decision Process (MDP) that takes into account a variety of energy-related aspects, including the currently available energy and that harvestable in the foreseeable future. Solution of the MDP is provided by a computationally light heuristic based on a simple threshold policy, thus obtaining further computational energy savings. The performance of G-WHARP is evaluated via GreenCastalia simulations, where we accurately model wake-up radios, harvestable energy, and the computational power needed to solve the MDP. Key network and system parameters are varied, including the source of harvestable energy, the network density, wake-up radio data rate and data traffic. We also compare the performance of G-WHARP to that of two state-of-the-art data forwarding strategies, namely GreenRoutes and CTP-WUR. Results show that G-WHARP limits energy expenditures while achieving low end-to-end latency and high packet delivery ratio. Particularly, it consumes up to 34% and 59% less energy than CTP-WUR and GreenRoutes, respectively.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1098/rspa.2021.0618
