arXiv research

Revisits VIC method to correct intrinsic reward bias in stochastic environments.

A new multi-objective RL framework improves intrinsic exploration performance.

A new reinforcement learning method uses mutual information to encourage agents to control their environment.

Exploration in sparse reward reinforcement learning remains an open challenge. Many state-of-the-art methods use intrinsic motivation to complement the sparse extrinsic reward signal, giving the agent more opportunities to receive feedback during exploration. Commonly these signals are added as bonus rewards, which res…

We study the intrinsic structure of parametric minimal discs in metric spaces admitting a quadratic isoperimetric inequality. We associate to each minimal disc a compact, geodesic metric space whose geometric, topological, and analytic properties are controlled by the isoperimetric inequality. Its geometry can be used …

New estimators for intrinsic dimension and Wasserstein distance improve OT accuracy.

In this contribution we present an intrinsic description of time-variant Port Hamiltonian systems as they appear in modeling and control theory. This formulation is based on the splitting of the state bundle and the use of appropriate covariant derivatives, which guarantees that the structure of the equations is invari…

Paper proposes a new method to optimize robot body structure and control policy.

We present an intrinsic formulation of the kinematic problem of two $n-$dimensional manifolds rolling one on another without twisting or slipping. We determine the configuration space of the system, which is an $\frac{n(n+3)}2-$dimensional manifold. The conditions of no-twisting and no-slipping are decoded by means of …

Solving tasks with sparse rewards is one of the most important challenges in reinforcement learning. In the single-agent setting, this challenge is addressed by introducing intrinsic rewards that motivate agents to explore unseen regions of their state spaces; however, applying these techniques naively to the multi-age…

A control system $\dot{q} = f(q,u)$ is said to be trivializable if there exists local coordinates in which the system is feedback equivalent to a control system of the form $\dot{q} = f(u)$. In this paper we characterize trivializable control systems and control systems for which, up to a feedback transformation, $f$ a…

Study shows tori metrics converging to flat under specific conditions.

New RL approach uses future state and action visitation measures for better exploration.

This work discusses a closed-loop control strategy for complex systems utilizing scarce and streaming data. A discrete embedding space is first built using hash functions applied to the sensor measurements from which a Markov process model is derived, approximating the complex system's dynamics. A control strategy is t…

The paper introduces a new intrinsic reward method for exploration in reinforcement learning.

Universal AI seeks high-optionality states through empowerment and curiosity.

Combining model-based and model-free deep reinforcement learning has shown great promise for improving sample efficiency on complex control tasks while still retaining high performance. Incorporating imagination is a recent effort in this direction inspired by human mental simulation of motor behavior. We propose a lea…

This paper improves robot grasping by integrating meta-control and latent-space imagination.

Study on rolling Stiefel manifolds with specific metrics.

The symmetric product of vector fields on a manifold arises when one studies the controllability of certain classes of mechanical control systems. A geometric description of the symmetric product is provided using parallel transport, along the lines of the flow interpretation of the Lie bracket. This geometric interpre…

The paper is devoted to the local classification of generic control-affine systems on an n-dimensional manifold with scalar input for any n>3 or with two inputs for n=4 and n=5, up to state-feedback transformations, preserving the affine structure. First using the Poincare series of moduli numbers we introduce the intr…

A well known question in differential geometry is to control the constant in isoperimetric inequality by intrinsic curvature conditions. In dimension 2, the constant can be controlled by the integral of the positive part of the Gaussian curvature. In this paper, we showed that on simply connected conformal flat manifol…

SHADOWCAST generates graphs with user-specified attributes.

Low-dimensional structure in images helps deep learning models generalize better.

The paper compares numerical schemes for nonholonomic systems using retraction maps.

We prove that for the mean curvature flow of two-convex hypersurfaces the intrinsic diameter stays uniformly controlled as one approaches the first singular time. We also derive sharp $L^{n-1}$-estimates for the regularity scale of the level set flow with two-convex initial data. Our proof relies on a detailed analysis…

In this paper, we describe a geometric setting for higher-order lagrangian problems on Lie groups. Using left-trivialization of the higher-order tangent bundle of a Lie group and an adaptation of the classical Skinner-Rusk formalism, we deduce an intrinsic framework for this type of dynamical systems. Interesting appli…

Study of Brown--York mass for four-dimensional asymptotically flat manifolds.

Novelty search in low-dimensional space improves sample efficiency in exploration tasks.

Exploration in environments with continuous control and sparse rewards remains a key challenge in reinforcement learning (RL). Recently, surprise has been used as an intrinsic reward that encourages systematic and efficient exploration. We introduce a new definition of surprise and its RL implementation named Variation…

In [7] Klainerman introduced the hyperboloidal method to prove the global existence results for nonlinear Klein-Gordon equations by using commuting vector fields. In this paper, we extend the hyperboloidal method from Minkowski space to Lorentzian spacetimes. This approach is developed in [14] for proving, under the ma…

A new method monitors unstructured 3D shapes without registration.

It has been established that diverse behaviors spanning the controllable subspace of an Markov decision process can be trained by rewarding a policy for being distinguishable from other policies \citep{gregor2016variational, eysenbach2018diversity, warde2018unsupervised}. However, one limitation of this formulation is …

EXOC framework uses auxiliary variables for counterfactual fairness in machine learning.

Deep neural networks solve stochastic control problems with delay.

Survey explores geometric aspects of policy optimization in control systems.

Federated framework learns causal states to predict counterfactuals without centralizing data.

We study deformations of Lie groupoids by means of the cohomology which controls them. This cohomology turns out to provide an intrinsic model for the cohomology of a Lie groupoid with values in its adjoint representation. We prove several fundamental properties of the deformation cohomology including Morita invariance…

The famous Nash embedding theorem published in 1956 was aiming for the opportunity to use extrinsic help in the study of (intrinsic) Riemannian geometry, if Riemannian manifolds could be regarded as Riemannian submanifolds. However, this hope had not been materialized yet according to \cite{G}. The main reason for this…

A new method evolves point clouds using B-splines for smooth surfaces.

Study on NNs for forecasting time series with novel control variable combinations.

Develops probabilistic safety regions for scalable classifiers.

We investigate the adaptive robust control framework for portfolio optimization and loss-based hedging under drift and volatility uncertainty. Adaptive robust problems offer many advantages but require handling a double optimization problem (infimum over market measures, supremum over the control) at each instance. Mor…

New budget quantifies drift in closed-loop learning, improving reproducibility.

Robot learns multiple tasks hierarchically by transferring knowledge.

AdaRL improves robust RL by adaptively adjusting policy complexity.

In the present paper we give a historical account -ranging from classical to modern results- of the problem of rolling two Riemannian manifolds one on the other, with the restrictions that they cannot instantaneously slip or spin one with respect to the other. On the way we show how this problem has profited from the d…

Introduces intrinsic Hopf-Lax semigroup linking to intrinsic slope.