Heterogeneous multi agent shared control with reinforcement learning based authority arbitration for semi closed loop EEG based shared autonomy

EEG-based shared autonomy integrates neural commands with autonomous assistance to improve control reliability under uncertain brain signals. However, EEG commands may exhibit varying confidence levels and conflict with autonomous actions, making dynamic authority allocation a key challenge. This study proposes a conflict-aware, functionally heterogeneous multi-agent shared-control framework with reinforcement-learning-based authority arbitration for semi-closed-loop EEG-based shared autonomy. Rather than jointly training all agents through reinforcement learning, the framework uses a pretrained EEG intention agent, a fixed autonomous control agent, and a reinforcement-learning-based authority-arbitration agent with distinct information sources and decision roles. The EEG intention agent outputs the estimated user action and its confidence score, while the autonomous control agent generates an environment-aware assistive action. The authority-arbitration agent then learns the control mode and user-authority ratio from EEG confidence, action disagreement, task state, and a shared task reward. A feedback-guided mechanism further adjusts subsequent authority allocation according to completed task-state transitions. The method was evaluated on the Kaya MI-BCI Dataset and Mobile BCI Dataset using recorded EEG data in a semi-closed-loop virtual-navigation environment rather than real-time human–BCI interaction. It achieved TSR/CRR values of 91.7%/88.4% and 87.8%/84.5%, respectively. Compared with EEG-TD3 Multi-agent Copilot, the method improved TSR by 1.1 and 1.3% points and CRR by 1.5 and 1.8% points on the two datasets, respectively. It also exceeded the matched single-agent RL arbitration baseline in both TSR and CRR. Feature-space, feature-importance, failure-case, and ablation analyses further examined the roles of EEG confidence, action disagreement, functional decomposition, and feedback-guided adjustment. The results support the feasibility of adaptive authority arbitration within the evaluated semi-closed-loop protocol and provide preliminary evidence for subsequent online closed-loop validation.

Authors

Institutions

Publication Details

Journal
Discover Artificial Intelligence
Published
2026-11-20
DOI
https://doi.org/10.1007/s44163-026-02792-x
Primary Topic
EEG and Brain-Computer Interfaces
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Heterogeneous multi agent shared control with reinforcement learning based authority arbitration for semi closed loop EEG based shared autonomy

Ao Zhang
3 links
Discover Artificial Intelligence
EEG and Brain-Computer Interfaces
article

Heterogeneous multi agent shared control with reinforcement learning based authority arbitration for semi closed loop EEG based shared autonomy

Ao Zhang
article en

Abstract

EEG-based shared autonomy integrates neural commands with autonomous assistance to improve control reliability under uncertain brain signals. However, EEG commands may exhibit varying confidence levels and conflict with autonomous actions, making dynamic authority allocation a key challenge. This study proposes a conflict-aware, functionally heterogeneous multi-agent shared-control framework with reinforcement-learning-based authority arbitration for semi-closed-loop EEG-based shared autonomy. Rather than jointly training all agents through reinforcement learning, the framework uses a pretrained EEG intention agent, a fixed autonomous control agent, and a reinforcement-learning-based authority-arbitration agent with distinct information sources and decision roles. The EEG intention agent outputs the estimated user action and its confidence score, while the autonomous control agent generates an environment-aware assistive action. The authority-arbitration agent then learns the control mode and user-authority ratio from EEG confidence, action disagreement, task state, and a shared task reward. A feedback-guided mechanism further adjusts subsequent authority allocation according to completed task-state transitions. The method was evaluated on the Kaya MI-BCI Dataset and Mobile BCI Dataset using recorded EEG data in a semi-closed-loop virtual-navigation environment rather than real-time human–BCI interaction. It achieved TSR/CRR values of 91.7%/88.4% and 87.8%/84.5%, respectively. Compared with EEG-TD3 Multi-agent Copilot, the method improved TSR by 1.1 and 1.3% points and CRR by 1.5 and 1.8% points on the two datasets, respectively. It also exceeded the matched single-agent RL arbitration baseline in both TSR and CRR. Feature-space, feature-importance, failure-case, and ablation analyses further examined the roles of EEG confidence, action disagreement, functional decomposition, and feedback-guided adjustment. The results support the feasibility of adaptive authority arbitration within the evaluated semi-closed-loop protocol and provide preliminary evidence for subsequent online closed-loop validation.

Discover Artificial IntelligenceVol. 6(1)
Shenyang Institute of Science and Technology (CN)
EEG and Brain-Computer Interfaces
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon