XAI-N: Sensor-based Robot Navigation using Expert Policies and Decision Trees

Roth, Aaron M.; Liang, Jing; Manocha, Dinesh

doi:10.1109/iros51168.2021.9636759

Cited by 8 publications

(6 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Once users understand that reasoning, they can use the information to modify, debug, or diagnose the system. XAI can also show users existing policies learned by an algorithm, which the users can then analyze, interpret, and change [15]. Even specific kinds of concerns can be addressed through this type of XAI without having to retrain the system.…”

Section: Explainable Artificial Intelligence a Background Of Xaimentioning

confidence: 99%

“…Roth et al [15] developed "a novel sensor-based learning navigation algorithm to compute a collision-free trajectory for a robot in dense and dynamic environments with moving obstacles or targets." This method relies on an expert policy based on deep reinforcement learning; it is also integrated with a policy extraction technique which generates the learned policies as a decision tree format.…”

Section: Related Researchmentioning

confidence: 99%

“…This method relies on an expert policy based on deep reinforcement learning; it is also integrated with a policy extraction technique which generates the learned policies as a decision tree format. The decision trees can be used to understand, analyze, and change policies as needed to improve performance [15]. The rule extraction and analysis can be used to solve problems, for example, to decrease immobilization frequency, improve smoothness, manage cases which lead to failure, or increase reliability, all without the need to retrain the algorithm.…”

Section: Related Researchmentioning

confidence: 99%

“…Demonstrates a new tool to clearly explain the production and energy consumption of a factory, facilitating anomaly and inefficiency detection; includes real-world industrial factory implementations Does not include detection or adaptation of data shift or concept drift; model requires a priori knowledge about system hierarchy and factory processes [60] Healthcare systems, network relaying schemes XAI implementation for optimal communication in healthcare system non-terrestrial networks using XAI-based cooperative relaying scheme Does not include dynamic signal-to-noise ratio; current implementation has a limited number of cooperative nodes [15] Robot navigation in dynamic environments, XAI Sensor-based robot navigation; use of decision trees, expert policies, and policy extraction to modify existing policies; modification of learning algorithm without retraining Limited problem-solving applications are tested (blocking, oscillation, freezing); obstacles cannot be moving faster than the maximum speed of the robot [17] XAI in manufacturing, review of XAI Provides a thorough review of XAI and desiderata, specific evaluation measures, list and classification of XAI techniques, manufacturing use cases, and applications in manufacturing Does not discuss cyber-resilience adapt as the CPS changes over time and to reason about possible anomalous behavior by the system. However, some researchers argue that existing XAI methods used for CPS are not effective in providing understandable explanations for time-series information [6], [67].…”

Section: B the Current Status Of Xai For Cyber-physical Systemsmentioning

confidence: 99%

See 3 more Smart Citations

Explainable AI for Cyber-Physical Systems: Issues and Challenges

Hoenig,

Roy,

Acquaah

et al. 2024

IEEE Access

View full text Add to dashboard Cite

Artificial intelligence and cyber-physical systems (CPS) are two of the key technologies of the future that are enabling major global shifts. However, most of the current implementations of AI in CPS are not explainable, which creates serious problems in ethical, legal, regulatory, and other domains. Therefore, it is necessary for explainable artificial intelligence (XAI) to be integrated with cyber-physical systems to meet the vital needs for control, fairness, accountability, safety, cyber-resilience, and cybersecurity. The goal of this review is to demonstrate the need, benefits, challenges, and implementation of XAI for CPS. We review the existing literature about XAI and CPS, discuss the current state of the art, examine applications in different domains, and make recommendations for future research directions. To the best of our knowledge, this is the first peer-reviewed academic article to provide a comprehensive review of general XAI for CPS. We also contribute new research ideas including development of multisensory explanations and outputs for these systems, application of XAI to CPS to decrease occupational burnout and increase employee engagement, and enumeration of the multidisciplinary goals and benefits of XAI as applied to cyber-physical systems.

show abstract

Section: Explainable Artificial Intelligence a Background Of Xaimentioning

confidence: 99%

Section: Related Researchmentioning

confidence: 99%

Section: Related Researchmentioning

confidence: 99%

Section: B the Current Status Of Xai For Cyber-physical Systemsmentioning

confidence: 99%

See 2 more Smart Citations

Explainable AI for Cyber-Physical Systems: Issues and Challenges

Hoenig,

Roy,

Acquaah

et al. 2024

IEEE Access

View full text Add to dashboard Cite

show abstract

“…Learning methods are increasingly being used for robot navigation in indoor and outdoor scenes. These include deep reinforcement learning (DRL) methods [1], [2], [3], [4], [5], [6], [7] and learning from demonstration [8], [9]. These methods have been evaluated in real-world scenarios, including obstacle avoidance, dynamic scenes, and uneven terrains.…”

Section: Introductionmentioning

confidence: 99%

MSVIPER: Improved Policy Distillation for Reinforcement-Learning-Based Robot Navigation

Roth¹,

Ji²,

Sriram³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

We present Multiple Scenario Verifiable Reinforcement Learning via Policy Extraction (MSVIPER), a new method for policy distillation to decision trees for improved robot navigation. MSVIPER learns an "expert" policy using any Reinforcement Learning (RL) technique involving learning a state-action mapping and then uses imitation learning to learn a decision-tree policy from it. We demonstrate that MSVIPER results in efficient decision trees and can accurately mimic the behavior of the expert policy. Moreover, we present efficient policy distillation and tree-modification techniques that take advantage of the decision tree structure to allow improvements to a policy without retraining. We use our approach to improve the performance of RL-based robot navigation algorithms for indoor and outdoor scenes. We demonstrate the benefits in terms of reduced freezing and oscillation behaviors (by up to 95% reduction) for mobile robots navigating among dynamic obstacles and reduced vibrations and oscillation (by up to 17%) for outdoor robot navigation on complex, uneven terrains.

show abstract