Human Behavior Modeling via Identification of Task Objective and Variability

TL;DR

The paper addresses predicting human control in human-automation loops by separating behavior into a task objective and intrinsic variability. It combines inverse optimal control to recover an explainable quadratic objective $(\mathbf{Q},\mathbf{R},\mathbf{S})$ with a DARE-based $\mathbf{P}$ and a probabilistic Gaussian mixture model to capture state-dependent variability $p(\mathbf{w}_k|\mathbf{x}_k)$. The authors introduce task-parameterized variability to generalize across environments and demonstrate their approach on a quadrotor landing task, showing improved trajectory prediction and confidence bounds compared to IOC-only and GMR-only baselines. The results indicate enhanced interpretability of human strategies, data efficiency, and robust prediction suitable for shared-control and shared-automation applications. Overall, the framework provides a principled, probabilistic, and data-efficient pathway to model and predict human behavior in dynamic control tasks.

Abstract

Human behavior modeling is important for the design and implementation of human-automation interactive control systems. In this context, human behavior refers to a human's control input to systems. We propose a novel method for human behavior modeling that uses human demonstrations for a given task to infer the unknown task objective and the variability. The task objective represents the human's intent or desire. It can be inferred by the inverse optimal control and improve the understanding of human behavior by providing an explainable objective function behind the given human behavior. Meanwhile, the variability denotes the intrinsic uncertainty in human behavior. It can be described by a Gaussian mixture model and capture the uncertainty in human behavior which cannot be encoded by the task objective. The proposed method can improve the prediction accuracy of human behavior by leveraging both task objective and variability. The proposed method is demonstrated through human-subject experiments using an illustrative quadrotor remote control example.

Figures (11)

• Figure 1: The variability in a quadrotor landing scenario; a human operator demonstrated the landing scenario for 30 times ($M=30$) with the same initial condition, but the demonstrated trajectories are varying for each trial due to the variability.
• Figure 2: A block scheme of the proposed method.
• Figure 3: (Left) Schematic diagram of the quadrotor landing simulator. (Right) Physical configuration of the testbed with a human operator.
• Figure 4: Collected trajectories under (Left) CS1 and (Right) CS2.
• Figure 5: An inferred task objective $\{ \hat{\mathbf{Q}},\hat{\mathbf{R}} \}_s, s \in \{1,2\}$ with normalization. The red-box represents the third diagonal element of $\hat{\mathbf{Q}}_s$ (a quadratic cost element on the attitude $\phi$).

Theorems & Definitions (3)

• Remark 1
• Remark 2
• Remark 3