arXiv research

Max entropy exploration guides reinforcement learning agents to pursue achievable goals.

Paper proposes Imitative Models combining IL and planning for flexible goal achievement.

A novel RL objective and prioritization framework improve performance and sample-efficiency in multi-goal tasks.

LEXA learns to discover and achieve goals in unseen environments.

Deep RL optimizes goal-based investing strategies.

Algorithm learns goals without rewards, controls environments.

PlanGAN uses GANs to plan efficient trajectories for multi-goal tasks in sparse reward environments.

Sparse reward problems are one of the biggest challenges in Reinforcement Learning. Goal-directed tasks are one such sparse reward problems where a reward signal is received only when the goal is reached. One promising way to train an agent to perform goal-directed tasks is to use Hindsight Learning approaches. In thes…

The paper optimizes insurance purchases for financial goals.

Generative neural nets learn deep policies conditioned on goals.

OptiGAN uses GAN and RL to optimize sequence generation for specific goals.

HGG generates goals to improve sample efficiency in robotic tasks.

We extend the classical risk minimization model with scalar risk measures to the general case of set-valued risk measures. The problem we obtain is a set-valued optimization model and we propose a goal programming-based approach with satisfaction function to obtain a solution which represents the best compromise betwee…

For an autonomous agent to fulfill a wide range of user-specified goals at test time, it must be able to learn broadly applicable and general-purpose skill repertoires. Furthermore, to provide the requisite level of generality, these skills must handle raw sensory input such as images. In this paper, we propose an algo…

The paper improves HER by prioritizing virtual goals and removing misleading samples.

UC-SSP algorithm tackles exploration in stochastic shortest path problems without loop-free assumption.

This work learns a distance function for goal-conditioned RL without prior knowledge.

A variety of cooperative multi-agent control problems require agents to achieve individual goals while contributing to collective success. This multi-goal multi-agent setting poses difficulties for recent algorithms, which primarily target settings with a single global reward, due to two new challenges: efficient explo…

Eikonal-Constrained QRL improves goal-reaching in reinforcement learning.

Adapts Floyd-Warshall algorithm for RL to improve multi-goal task learning.

Automatically generates curricula for reinforcement learning agents.

This work improves imitation learning and goal-conditioned RL by estimating value densities.

The paper tackles online learning with two types of losses and shows it's impossible without certain assumptions.

ACTRCE uses natural language to improve reinforcement learning performance.

All-goals updating exploits the off-policy nature of Q-learning to update all possible goals an agent could have from each transition in the world, and was introduced into Reinforcement Learning (RL) by Kaelbling (1993). In prior work this was mostly explored in small-state RL problems that allowed tabular representati…

Robotic arm learns to manipulate a ball by choosing goals from learned experience.

AMIGo uses adversarial intrinsic goals to teach RL agents new skills.

C-Learning estimates reachability over time to solve multi-goal tasks.

Agents need world models to generalize multi-step tasks.

Study combines chit-chat and goal-oriented dialogue in fantasy games.

Gold Seeker uses reinforcement learning to select actions that maximize information gain.

New modeling approach for self-organizing complex systems.

Generating novel graph structures that optimize given objectives while obeying some given underlying rules is fundamental for chemistry, biology and social science research. This is especially important in the task of molecular graph generation, whose goal is to discover novel molecules with desired properties such as …

In this paper, we study a new learning paradigm for Neural Machine Translation (NMT). Instead of maximizing the likelihood of the human translation as in previous works, we minimize the distinction between human translation and the translation given by an NMT model. To achieve this goal, inspired by the recent success …

A key challenge in complex visuomotor control is learning abstract representations that are effective for specifying goals, planning, and generalization. To this end, we introduce universal planning networks (UPN). UPNs embed differentiable planning within a goal-directed policy. This planning computation unrolls a for…

The goals of this article are twofold : 1) to compute the conjugate locus of a geodesic that lies in the center of a simply connected, 2-step nilpotent Lie group with a left invariant metric 2) compare the isometry types of two such nilpotent Lie groups whose conjugate loci for central geodesics are "the same" in a sui…

New methods improve accuracy in predicting complex systems.

Goals for reinforcement learning problems are typically defined through hand-specified rewards. To design such problems, developers of learning algorithms must inherently be aware of what the task goals are, yet we often require agents to discover them on their own without any supervision beyond these sparse rewards. W…

Method maps state space using landmarks for universal goal reaching.

A new method for robot manipulation tasks using imagined object goals.

The paper develops methods to predict the probability of achieving a user goal in a task, ensuring the system alerts when the probability falls below a threshold.

ADRL improves participant selection in MCS systems.

Improves reinforcement learning for complex tasks with sparse feedback.

We present an active learning architecture that allows a robot to actively learn which data collection strategy is most efficient for acquiring motor skills to achieve multiple outcomes, and generalise over its experience to achieve new outcomes. The robot explores its environment both via interactive learning and goal…

CDP improves RL performance and sample-efficiency by prioritizing rare goal states.

We study the problem of efficient PAC active learning of homogeneous linear classifiers (halfspaces) in $\mathbb{R}^d$, where the goal is to learn a halfspace with low error using as few label queries as possible. Under the extra assumption that there is a $t$-sparse halfspace that performs well on the data ($t \ll d$)…

GoBOED optimizes experiments for specific decision-making objectives, improving downstream outcomes.

The paper proposes a method to transfer skills between tasks using disentangled latent policies.