Towards Deviation-Robust Agent Navigation via Perturbation-Aware Contrastive Learning.

world, by requiring them to learn towards deviation-robust navigation. ... For encouraging the agent to well capture the difference brought by perturbation and adapt to both perturbation-free and perturbation-based environments, a perturbation-aware contrastive learning mechanism ... E CONCLUSION This paper proposes Progressive Perturbation-aware Contrastive Learning (PROPER) for training deviation-robust VLN agents, which introduces a simple yet effective path perturbation scheme ...

doi:10.1109/tpami.2023.3273594 pmid:37155380 fatcat:grtlop3xtzgfdijgmx3kmxibme

Multiple Versions

Adversarial attacks are valuable for providing insights into the blind-spots of deep learning models and help improve their robustness. ... Existing work on adversarial attacks have mainly focused on static scenes; however, it remains unclear whether such attacks are effective against embodied agents, which could navigate and interact with ... The results in Table 3 support the fact that training on our adversarial perturbations can improve the agent robustness towards some types of noises (i.e., higher QA accuracy, and lower d T ). ...

arXiv:2005.09161v3 fatcat:dwmgy57qdzeu5ptifofsmthtfa

Multiple Versions

Reinforcement learning agents are susceptible to evasion attacks during deployment. ... In single-agent environments, these attacks can occur through imperceptible perturbations injected into the inputs of the victim policy network. ... Navigation Task. ...

arXiv:2305.02605v3 fatcat:ajyuqvlg5vd2ncx6yew7sibbs4

Multiple Versions

environmental novelties through Cooperative Multi-Agent Reinforcement Learning (CMARL) algorithms. ... However, this cooperative learning process is susceptible to adversarial poisoning attacks, as highlighted by contemporary research. ... Essentially, a protocol adheres to LDP if any two navigational Q-values are perturbed to the same value with similar probabilities. ...

doi:10.1145/3643137 fatcat:dxp6dtl2efdrrmqenjxa6njjly

We trained deep reinforcement learning agents using neural architectures with various degrees of modularity in a partially observable navigation task. ... Artificial reinforcement learning agents that perform well in training tasks typically perform worse than animals in novel tasks. ... This result is expected, given the structure of the novel tasks: agents must be aware of novel gains or perturbations via optic flow, since their internal model for prediction is outdated in novel tasks ...

doi:10.1101/2022.12.07.519515 fatcat:j662n73qzvdarjia22ch3ae5r4

Results show that the resultant neural mass model allows the agent to reproduce the navigational patterns of a rodent in an open field. ... Moreover, when exploring the robustness of our model in a dynamically changing environment, the synthetic agent pursues the stability of the self, being its internal states dependent on environmental opportunities ... Once external perturbations produce a deviation from the desirable range, a homeostatic error arises, driving a proportional error-correcting response to restore balance in the system. ...

doi:10.3389/frobt.2022.1052998 pmid:36530500 pmcid:PMC9755511 fatcat:vfgwv44x2ncw3awbvgrtpjuj3q

DOAJ

In this study, we propose an uncertainty-aware distributional offline RL method to simultaneously address both epistemic uncertainty and environmental stochasticity. ... Offline reinforcement learning (RL) presents distinct challenges as it relies solely on observational data. ... In contrast, UDAC aims to learn to model the behavior policy from the datasets directly via denoising, which has better generalizability. ...

arXiv:2403.17646v1 fatcat:tu3oidxeurb5tpq462xewm4h64

In this paper, we present an adaptive neural network-based system for co-robot control that employs exploratory learning to achieve the coordinated motor planning needed to navigate toward, reach for, ... Collaborative robots, or co-robots, are semi-autonomous robotic agents designed to work alongside humans in shared workspaces. ... Acknowledgments This work was supported by the Center of Excellence for Learning in Education, Science, and Technology, a National Science Foundation Science of Learning Center (NSF SMA-0835976). ...

doi:10.3389/fnbot.2015.00007 pmid:26257640 pmcid:PMC4511843 fatcat:rws3ea56cffxnpintwiyas4nfe

DOAJ

The identified critical enabling techniques and technologies for adaptive, informative and reconfigurable operations of unmanned swarm systems are robust static sensor network design, mobile sensor tasking ... Swarms of unmanned vehicles (air and ground) can increase the efficiency and effectiveness of military and law enforcement operations by enhancing situational awareness and allowing the persistent monitoring ... To evaluate the robustness against navigation errors, the true sensor position and orientation are perturbed by band limited Gaussian processes. ...

doi:10.3390/drones6040094 fatcat:tarq7frvrvflbfspe3qxplo534

DOAJ

Towards a more general solution, we formulate the multi-set robustness problem to learn a policy robust to different perturbation sets. ... Reinforcement learning (RL) agents need to be robust to variations in safety-critical environments. ... Since the optimal robust policy varies across different perturbation sets, we incorporate the uncertainty set as contextual information to the agent and learn a generalized set-conditioned policy. ...

arXiv:2202.07013v2 fatcat:dh6lnzrvinagfhszj5rlgm3kzm

Multiple Versions

Agents are tasked to navigate towards a target goal by executing a set of low-level actions, following a series of natural language instructions. ... This benchmark provides valuable insight into the robustness of VLN systems in continuous environments. ... In future work, we will investigate error-aware policy learning to improve the navigation performance in the VLN-CE task. ...

arXiv:2403.10700v1 fatcat:dvnvpy6wnnd6bgdclhsp2ebize

In multi-agent navigation, agents need to move towards their goal locations while avoiding collisions with other agents and static obstacles, often without communication with each other. ... We accomplish this by formulating the multi-agent navigation problem as an action-selection problem, and propose an approach, ALAN, that allows agents to compute time-efficient and collision-free motions ... Offline learning has significant limitations, which arise from the need to train the agents before the environment is known. In contrast, the main part of our work is an online learning approach. ...

arXiv:1710.04296v1 fatcat:4xsmrhlyurc3nagmgbbe4opiym

This essay describes a general approach to building perturbation-tolerant autonomous systems, based on the conviction that artificial agents should be able to notice when something is amiss, assess the ... This basic strategy of self-guided learning is termed the metacognitive loop; it involves the system monitoring, reasoning about, and, when necessary, altering its own decision-making components. ... The agent exhibits better behaviour while training, and also learns more quickly to navigate effectively [28] . ...

doi:10.1093/logcom/exh034 fatcat:pfifnzq3avdrdm57ovj5dbvkcy

In this article, we propose a scalable lifelong RL method that dynamically expands the network capacity to accommodate new knowledge while preventing past memories from being perturbed. ... While reinforcement learning (RL) algorithms are achieving state-of-the-art performance in various challenging tasks, they can easily encounter catastrophic forgetting or interference when faced with lifelong ... Robust Prior via Domain Randomization We formulate a mixture of task models for performing lifelong learning adaptation in the face of an infinite stream of incoming data. ...

doi:10.1109/tcyb.2022.3170485 pmid:35580095 fatcat:5wxhflq25vbthazi25jesbmhim

Multiple Versions

Reinforcement Learning (RL) controllers have proved to effectively tackle the dual objectives of path following and collision avoidance. ... Compared to the introduced RL algorithms, the results show that the Proximal Policy Optimization (PPO) algorithm exhibits superior robustness to changes in the environment complexity, the reward function ... The time consumption (simulation steps) per episode measures whether the agent can make quick decisions, navigate at high speeds, and take the shortest deviating path, in contrast to navigating far away ...

doi:10.3389/frobt.2021.738113 pmid:34589522 pmcid:PMC8473616 fatcat:dprzo5pjmfh2nnotmwrk6at3xa

DOAJ

Towards Deviation-Robust Agent Navigation via Perturbation-Aware Contrastive Learning

Preserved Fulltext

Spatiotemporal Attacks for Embodied Agents [article]

Preserved Fulltext

Other Versions

Toward Evaluating Robustness of Reinforcement Learning with Adversarial Policy [article]

Preserved Fulltext

Other Versions

RAMPART: Reinforcing Autonomous Multi-agent Protection through Adversarial Resistance in Transportation

Preserved Fulltext

Inductive biases of neural networks for generalization in spatial navigation [article]

Preserved Fulltext

Drive competition underlies effective allostatic orchestration

Preserved Fulltext

Uncertainty-aware Distributional Offline Reinforcement Learning [article]

Preserved Fulltext

A neural network-based exploratory learning and motor planning system for co-robots

Preserved Fulltext

Autonomous Unmanned Heterogeneous Vehicles for Persistent Monitoring

Preserved Fulltext

Robust Policy Learning over Multiple Uncertainty Sets [article]

Preserved Fulltext

Other Versions

Mind the Error! Detection and Localization of Instruction Errors in Vision-and-Language Navigation [article]

Preserved Fulltext

ALAN: Adaptive Learning for Multi-Agent Navigation [article]

Preserved Fulltext

Logic, Self-awareness and Self-improvement: the Metacognitive Loop and the Problem of Brittleness

Preserved Fulltext

A Dirichlet Process Mixture of Robust Task Models for Scalable Lifelong Reinforcement Learning

Preserved Fulltext

Comparing Deep Reinforcement Learning Algorithms' Ability to Safely Navigate Challenging Waters

Preserved Fulltext