We discuss nonlinear model predictive control (NMPC) for multi-body dynamics via physics-informed machine learning methods. Physics-informed neural networks (PINNs) are a promising tool to approximate (partial) differential equations. PINNs are not suited for control tasks in their original form since they are not designed to handle variable control actions or variable initial values. We thus present the idea of enhancing PINNs by adding control actions and initial conditions as additional network inputs. The high-dimensional input space is subsequently reduced via a sampling strategy and a zero-hold assumption. This strategy enables the controller design based on a PINN as an approximation of the underlying system dynamics. The additional benefit is that the sensitivities are easily computed via automatic differentiation, thus leading to efficient gradient-based algorithms. Finally, we present our results using our PINN-based MPC to solve a tracking problem for a complex mechanical system, a multi-link manipulator.
相關內容
Nonprehensile manipulation involves long horizon underactuated object interactions and physical contact with different objects that can inherently introduce a high degree of uncertainty. In this work, we introduce a novel Real-to-Sim reward analysis technique, called Riemannian Motion Predictive Control (RMPC), to reliably imagine and predict the outcome of taking possible actions for a real robotic platform. Our proposed RMPC benefits from Riemannian motion policy and second order dynamic model to compute the acceleration command and control the robot at every location on the surface. Our approach creates a 3D object-level recomposed model of the real scene where we can simulate the effect of different trajectories. We produce a closed-loop controller to reactively push objects in a continuous action space. We evaluate the performance of our RMPC approach by conducting experiments on a real robot platform as well as simulation and compare against several baselines. We observe that RMPC is robust in cluttered as well as occluded environments and outperforms the baselines.
An Adaptive Framework for Reliable Trajectory Following in Changing-Contact Robot Manipulation Tasks
We describe a framework for changing-contact robot manipulation tasks that require the robot to make and break contacts with objects and surfaces. The discontinuous interaction dynamics of such tasks make it difficult to construct and use a single dynamics model or control strategy, and the highly non-linear nature of the dynamics during contact changes can be damaging to the robot and the objects. We present an adaptive control framework that enables the robot to incrementally learn to predict contact changes in a changing contact task, learn the interaction dynamics of the piece-wise continuous system, and provide smooth and accurate trajectory tracking using a task-space variable impedance controller. We experimentally compare the performance of our framework against that of representative control methods to establish that the adaptive control and incremental learning components of our framework are needed to achieve smooth control in the presence of discontinuous dynamics in changing-contact robot manipulation tasks.
We explore the possibility to use physics-informed neural networks to drastically accelerate the solution of ordinary differential-algebraic equations that govern the power system dynamics. When it comes to transient stability assessment, the traditionally applied methods either carry a significant computational burden, require model simplifications, or use overly conservative surrogate models. Conventional neural networks can circumvent these limitations but are faced with high demand of high-quality training datasets, while they ignore the underlying governing equations. Physics-informed neural networks are different: they incorporate the power system differential algebraic equations directly into the neural network training and drastically reduce the need for training data. This paper takes a deep dive into the performance of physics-informed neural networks for power system transient stability assessment. Introducing a new neural network training procedure to facilitate a thorough comparison, we explore how physics-informed neural networks compare with conventional differential-algebraic solvers and classical neural networks in terms of computation time, requirements in data, and prediction accuracy. We illustrate the findings on the Kundur two-area system, and assess the opportunities and challenges of physics-informed neural networks to serve as a transient stability analysis tool, highlighting possible pathways to further develop this method.
Soft robots are made of compliant and deformable materials and can perform tasks challenging for conventional rigid robots. The inherent compliance of soft robots makes them more suitable and adaptable for interactions with humans and the environment. However, this preeminence comes at a cost: their continuum nature makes it challenging to develop robust model-based control strategies. Specifically, an adaptive control approach addressing this challenge has not yet been applied to physical soft robotic arms. This work presents a reformulation of dynamics for a soft continuum manipulator using the Euler-Lagrange method. The proposed model eliminates the simplifying assumption made in previous works and provides a more accurate description of the robot's inertia. Based on our model, we introduce a task-space adaptive control scheme. This controller is robust against model parameter uncertainties and unknown input disturbances. The controller is implemented on a physical soft continuum arm. A series of experiments were carried out to validate the effectiveness of the controller in task-space trajectory tracking under different payloads. The controller outperforms the state-of-the-art method both in terms of accuracy and robustness. Moreover, the proposed model-based control design is flexible and can be generalized to any continuum robotic arm with an arbitrary number of continuum segments.
We propose a time-adaptive predictor/multi-corrector method to solve hyperbolic partial differential equations, based on the generalized-$\alpha$ scheme that provides user-control on the numerical dissipation and second-order accuracy in time. Our time adaptivity uses an error estimation that exploits the recursive structure of the variable updates. The predictor/multicorrector method explicitly updates the equation system but computes the residual of the system implicitly. We analyze the method's stability and describe how to determine the parameters that ensure high-frequency dissipation and accurate low-frequency approximation. Subsequently, we solve a linear wave equation, followed by non-linear finite strain deformation problems with different boundary conditions. Thus, our method is a straightforward, stable and computationally efficient approach to simulate real-world engineering problems. Finally, to show the performance of our method, we provide several numerical examples in two and three dimensions. These challenging tests demonstrate that our predictor/multicorrector scheme dynamically adapts to sudden energy releases in the system, capturing impacts and boundary shocks. The method efficiently and stably solves dynamic equations with consistent and under-integrated mass matrices conserving the linear and angular momenta as well as the system's energy for long-integration times.
The incorporation of appropriate inductive bias plays a critical role in learning dynamics from data. A growing body of work has been exploring ways to enforce energy conservation in the learned dynamics by encoding Lagrangian or Hamiltonian dynamics into the neural network architecture. These existing approaches are based on differential equations, which do not allow discontinuity in the states and thereby limit the class of systems one can learn. However, in reality, most physical systems, such as legged robots and robotic manipulators, involve contacts and collisions, which introduce discontinuities in the states. In this paper, we introduce a differentiable contact model, which can capture contact mechanics: frictionless/frictional, as well as elastic/inelastic. This model can also accommodate inequality constraints, such as limits on the joint angles. The proposed contact model extends the scope of Lagrangian and Hamiltonian neural networks by allowing simultaneous learning of contact and system properties. We demonstrate this framework on a series of challenging 2D and 3D physical systems with different coefficients of restitution and friction. The learned dynamics can be used as a differentiable physics simulator for downstream gradient-based optimization tasks, such as planning and control.
Humans have a strong intuitive understanding of the 3D environment around us. The mental model of the physics in our brain applies to objects of different materials and enables us to perform a wide range of manipulation tasks that are far beyond the reach of current robots. In this work, we desire to learn models for dynamic 3D scenes purely from 2D visual observations. Our model combines Neural Radiance Fields (NeRF) and time contrastive learning with an autoencoding framework, which learns viewpoint-invariant 3D-aware scene representations. We show that a dynamics model, constructed over the learned representation space, enables visuomotor control for challenging manipulation tasks involving both rigid bodies and fluids, where the target is specified in a viewpoint different from what the robot operates on. When coupled with an auto-decoding framework, it can even support goal specification from camera viewpoints that are outside the training distribution. We further demonstrate the richness of the learned 3D dynamics model by performing future prediction and novel view synthesis. Finally, we provide detailed ablation studies regarding different system designs and qualitative analysis of the learned representations.
We present an approach to learn an object-centric forward model, and show that this allows us to plan for sequences of actions to achieve distant desired goals. We propose to model a scene as a collection of objects, each with an explicit spatial location and implicit visual feature, and learn to model the effects of actions using random interaction data. Our model allows capturing the robot-object and object-object interactions, and leads to more sample-efficient and accurate predictions. We show that this learned model can be leveraged to search for action sequences that lead to desired goal configurations, and that in conjunction with a learned correction module, this allows for robust closed loop execution. We present experiments both in simulation and the real world, and show that our approach improves over alternate implicit or pixel-space forward models. Please see our project page (//judyye.github.io/ocmpc/) for result videos.
Neural Network Based Reinforcement Learning for Audio-Visual Gaze Control in Human-Robot Interaction
This paper introduces a novel neural network-based reinforcement learning approach for robot gaze control. Our approach enables a robot to learn and to adapt its gaze control strategy for human-robot interaction neither with the use of external sensors nor with human supervision. The robot learns to focus its attention onto groups of people from its own audio-visual experiences, independently of the number of people, of their positions and of their physical appearances. In particular, we use a recurrent neural network architecture in combination with Q-learning to find an optimal action-selection policy; we pre-train the network using a simulated environment that mimics realistic scenarios that involve speaking/silent participants, thus avoiding the need of tedious sessions of a robot interacting with people. Our experimental evaluation suggests that the proposed method is robust against parameter estimation, i.e. the parameter values yielded by the method do not have a decisive impact on the performance. The best results are obtained when both audio and visual information is jointly used. Experiments with the Nao robot indicate that our framework is a step forward towards the autonomous learning of socially acceptable gaze behavior.
Although reinforcement learning methods can achieve impressive results in simulation, the real world presents two major challenges: generating samples is exceedingly expensive, and unexpected perturbations can cause proficient but narrowly-learned policies to fail at test time. In this work, we propose to learn how to quickly and effectively adapt online to new situations as well as to perturbations. To enable sample-efficient meta-learning, we consider learning online adaptation in the context of model-based reinforcement learning. Our approach trains a global model such that, when combined with recent data, the model can be be rapidly adapted to the local context. Our experiments demonstrate that our approach can enable simulated agents to adapt their behavior online to novel terrains, to a crippled leg, and in highly-dynamic environments.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191