Towards Accountability: Providing Intelligible Explanations in Autonomous Driving

Omeiza, Daniel; Web, Helena; Jirotka, Marina; Kunze, Lars

doi:10.1109/iv48863.2021.9575917

Cited by 20 publications

(10 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Some authors advocate for interpretable approaches to explainability in AVs. For instance, Omeiza et al [19] proposed a social-technical approach to explainability and proposed an interpretable representation for explanations based on a combination of actions, observations, and road rules. Nahata et al [20] also applied an interpretable method to explain risk prediction models for autonomous vehicles.…”

Section: Related Workmentioning

confidence: 99%

From Spoken Thoughts to Automated Driving Commentary: Predicting and Explaining Intelligent Vehicles' Actions

Omeiza,

Anjomshoae,

Webb

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Commentary driving is a technique in which drivers verbalise their observations, assessments and intentions. By speaking out their thoughts, both learning and expert drivers are able to create a better understanding and awareness of their surroundings. In the intelligent vehicle context, automated driving commentary can provide intelligible explanations about driving actions, and thereby assist a driver or an end-user during driving operations in challenging and safety-critical scenarios. In this paper, we conducted a field study in which we deployed a research vehicle in an urban environment to obtain data. While collecting sensor data of the vehicle's surroundings, we obtained driving commentary from a driving instructor using the think-aloud protocol. We analysed the driving commentary and uncovered an explanation style; the driver first announces his observations, announces his plans, and then makes general remarks. He also made counterfactual comments. We successfully demonstrated how factual and counterfactual natural language explanations that follow this style could be automatically generated using a simple tree-based approach. Generated explanations for longitudinal actions (e.g., stop and move) were deemed more intelligible and plausible by human judges compared to lateral actions, such as lane changes. We discussed how our approach can be built on in the future to realise more robust and effective explainability for driver assistance as well as partial and conditional automation of driving functions.

show abstract

Section: Related Workmentioning

confidence: 99%

From Spoken Thoughts to Automated Driving Commentary: Predicting and Explaining Intelligent Vehicles' Actions

Omeiza,

Anjomshoae,

Webb

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…This would not be an optional extension for large-scale systems leveraging AI and instead it would be a critical requirement [37]. Accordingly, several research papers have established a solid definition for accountability by formulating different dimensions of it [6,37] and proposing approaches for realizing it in various domains [2,16,28,32].…”

Section: Accountability Problem Formulationmentioning

confidence: 99%

Accountability challenges of AI in smart grid services

Volkova

Patil

Javadi

et al. 2022

Proceedings of the Thirteenth ACM International Conference on Future Energy Systems

View full text Add to dashboard Cite

The rising importance and efficiency of Artificial Intelligence (AI) has led to AI being substantially deployed in various domains and its usage growing rapidly. For the smart grid, adopting AI-based methods has transformed established power system functions into more advanced analogs. However, the unstable operation of AIbased services may misguide operators to make incorrect operational decisions leading to substantial consequences. Accordingly, the rapidly growing market of AI-based services calls for raising awareness of accountability of these services. Accountability in grid services is getting even more challenging with respect to the wide availability of the AI-based methods implemented as libraries and the emergence of AI as a service.A key way forward is through smart grid operators monitoring their AI-based services to enable accountability. Accountable AI increases trustworthiness and accelerates acceptance of these services by society. Toward this, a conceptual representation of a monitoring regime for AI-based services is proposed and major challenges to realising it are identified. In addition, this work aims to draw more attention to AI-based services and the issues they raise, given AI increasing prominence and the general calls for the more responsible and accountable use of AI. CCS CONCEPTS• Security and privacy → Social aspects of security and privacy; • Applied computing → Investigation techniques.

show abstract

“…where M is equal to 6 as there are three predicted waypoints that have x,y coordinates for each. Similarly, we also use L1 loss to supervise navigational controls prediction as in (8). However, averaging is not needed as there is only one element for each output (steering and throttle).…”

Section: Trainingmentioning

confidence: 99%

“…Then, its performance on each task is calculated by a specific metric function. To evaluate waypoints and navigational controls prediction, we use mean absolute error (MAE) or L1 loss as in (7) and (8). Meanwhile, we compute intersection over union (IoU) as in (9) for evaluating the segmentation performance.…”

Section: E Evaluation and Scoringmentioning

confidence: 99%

“…All these benefits result in a better model performance even with smaller model size due to its compactness [1][2] [3]. To date, there have been a lot of works in the field of end-to-end autonomous driving, whether it is based on simulation [4] [5], or offline real world where the model predicts a set of driving records [6] [7], or online real world where the model is deployed for automated driving [8] [9]. One problem that remains in the online real world autonomous driving is that the model must deal with uncertain conditions including noise and inaccuracy of sensor measurement that affect the model performance [10] [11].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

DeepIPC: Deeply Integrated Perception and Control for an Autonomous Vehicle in Real Environments

Natan¹,

Miura²

2022

Preprint

View full text Add to dashboard Cite

We propose DeepIPC, an end-to-end multi-task model that handles both perception and control tasks in driving a mobile robot autonomously. The model consists of two main parts, perception and controller modules. The perception module takes RGB image and depth map to perform semantic segmentation and bird's eye view (BEV) semantic mapping along with providing their encoded features. Meanwhile, the controller module processes these features with the measurement of GNSS locations and angular speed to estimate waypoints that come with latent features. Then, two different agents are used to translate waypoints and latent features into a set of navigational controls to drive the robot. The model is evaluated by predicting driving records and performing automated driving under various conditions in the real environment. Based on the experimental results, DeepIPC achieves the best drivability and multi-task performance even with fewer parameters compared to the other models.

show abstract

Towards Accountability: Providing Intelligible Explanations in Autonomous Driving

Cited by 20 publications

References 24 publications

From Spoken Thoughts to Automated Driving Commentary: Predicting and Explaining Intelligent Vehicles' Actions

From Spoken Thoughts to Automated Driving Commentary: Predicting and Explaining Intelligent Vehicles' Actions

Accountability challenges of AI in smart grid services

DeepIPC: Deeply Integrated Perception and Control for an Autonomous Vehicle in Real Environments

Contact Info

Product

Resources

About