OMRON SINIC X Research Paper was Accepted as a Full Paper to the International Conference on Autonomous Agents and Multiagent Systems (AAMAS)

June 2, 2023

OMRON SINIC X Corporation (HQ: Bunkyo-ku, Tokyo; President and CEO: Masaki Suwa; hereinafter “OSX”) is pleased to announce that our research paper has been accepted to and presented at the International Conference on Autonomous Agents and Multiagent Systems 2023 (hereinafter “AAMAS 2023”) held in London on May 29.

AAMAS is the flagship annual conference of the non-profit International Foundation for Autonomous Agents and Multiagent Systems (IFAAMAS). More than 1,000 papers were submitted this year, with an acceptance rate of around 45% for poster presentation. The paper submitted by OSX was selected as a full paper (poster and oral presentation), which account for just 23.3% of all submissions.

Research Paper Selected:
Counterfactual Fairness Filter for Fair-Delay Multi-Robot Navigation
Authors:
　Hikaru Asano (The University of Tokyo; OSX Research Intern from March 2022)
　Ryo Yonetani (CyberAgent, Inc.; OSX at time of writing)
　Mai Nishimura (OSX)
　Tadashi Kozuno (OSX)

Background:

A decentralized multi-robot navigation problem considers a setting where multiple robots simultaneously and independently navigate to their destinations as quickly as possible (efficiency requirement) avoiding collisions with other robots and obstacles (safety requirement). It is a useful setting that models diverse applications such as food delivery, cab service, and disaster relief. However, the standard setting admits robot paths in which a few robots suffer huge delays while others experience no delay. Such a situation is unacceptable in applications such as those mentioned above as customers are treated unfairly.

Research overview:

Given the issue explained above, we proposed a new multi-robot navigation problem in which robots need to take into account not only efficiency and safety but also fairness. Formally, fairness in navigation is defined as the variance of delays among agents, where a delay of a robot is the difference between actual travel time, and a counterfactual travel time that the robot would need if there were no other agent.

Furthermore, we proposed an algorithm that leverages decentralized counterfactual inference1) for assuring fairness. Concretely, each robot has two systems and sends messages to its neighbors: one determining movements of the robot from the perspective of efficiency, and the other determing if the robot may move or must give way to other robots from the perspective of fairness given desired movements of other robots. The latter system makes decisions based on the degree to which delays for other robots would be reduced if the robot gave way. In this way, our proposed algorithm can balance the tradeoff between efficiency and fairness.

1) Counterfactual inference involves imagining a situation that differs from the current reality and predicting what would result from it. An example is the statement “if it had rained, we might have gotten wet”: the speaker is imagining a situation different from the reality in which it did not rain, and thinking about what would have happened if it did.

Results:

To test the effectiveness of our proposed method, we experimented with existing fairness-aware multi-agent reinforcement learning algorithm for comparison.

Black circles and squares: static obstacles. Non-filled circles: starting positions of each agent
Non-filled squares: goal positions of each agent correspond to the start positions with the same colors.
Figure 1: Example of multi-agent navigation

(a) Results for the problem with few obstacles (a) Results for the problem with few obstacles

(b) Results for the problem with many obstacles
Baseline method: Fair-Efficient Network (FEN)、Jiechuan Jiang and Zongqing Lu. 2019. Learning fairness in multi-agent systems. In Proceedings of the Annual Conference on Neural Information Processing Systems (NeurIPS). 13854–13865
Figure 2: Comparison of navigation calculation results for 16 different agents

Figure 2 (a) demonstrates that our method improves fairness in the variance of delays among agents by a factor of 2.3, while the baseline method suffers from unfairness in the navigation delays for problems with less obstacles. Figure 2 (b) also shows that our method still achieves higher fairness in the variance of delays compared to the baseline method for problems with more obstacles.

For more details, see the following article.
「 Counterfactual Fairness Filter for Fair-Delay Multi-Robot Navigation (AAMAS2023)」

OSX continue to develop value creation via technological innovation through collaboration with universities and external research institutions.

About OMRON SINIC X Corporation
OMRON SINIC X Corporation is a strategic subsidiary seeking to realize the “near future designs” that OMRON forecasts. Researchers with cutting-edge knowledge and experience across many technological domains, including AI, Robotics, IoT, and sensing, are affiliated with OSX, and with the aim of solving social issues, they are working to create near future designs by integrating innovative technologies with business models and strategies in technology and IP. The company will also accelerate the creation of near future designs through joint research with universities and external research institutions. For more details, please refer to https://www.omron.com/sinicx/en/

For press inquiries related to this release, please contact the following:: Tech Communications and Collaboration Promo Dept.
Strategy Division
Technology and Intellectual Property H.Q.
OMRON Corporation
Tel: +81-774-74-2010

Adobe Acrobat Reader is free software that lets you view and print Adobe Portable Document Format (PDF) files.