Bin Hu

Chen Xu, Zhenyu Lv, Tian Lan et al.

Although large language models (LLMs) hold significant promise in psychotherapy, their direct application in patient-facing scenarios raises ethical and safety concerns. Therefore, this work shifts towards developing an LLM as a supervisor to train real therapists. In addition to the privacy of clinical therapist training data, a fundamental contradiction complicates the training of therapeutic behaviors: clear feedback standards are necessary to ensure a controlled training system, yet there is no absolute "gold standard" for appropriate therapeutic behaviors in practice. In contrast, many common therapeutic mistakes are universal and identifiable, making them effective triggers for targeted feedback that can serve as clearer evidence. Motivated by this, we create a novel therapist-training paradigm: (1) guidelines for mistaken behaviors and targeted correction strategies are first established as standards; (2) a human-in-the-loop dialogue-feedback dataset is then constructed, where a mistake-prone agent intentionally makes standard mistakes during interviews naturally, and a supervisor agent locates and identifies mistakes and provides targeted feedback; (3) after fine-tuning on this dataset, the final supervisor model is provided for real therapist training. The detailed experimental results of automated, human and downstream assessments demonstrate that models fine-tuned on our dataset MATE, can provide high-quality feedback according to the clinical guideline, showing significant potential for the therapist training scenario.

Bo Wu, Bin Hu, Hai Lin

This paper considers an optimal task allocation problem for human robot collaboration in human robot systems with persistent tasks. Such human robot systems consist of human operators and intelligent robots collaborating with each other to accomplish complex tasks that cannot be done by either part alone. The system objective is to maximize the probability of successfully executing persistent tasks that are formulated as linear temporal logic specifications and minimize the average cost between consecutive visits of a particular proposition. This paper proposes to model the human robot collaboration under a framework with the composition of multiple Markov Decision Process (MDP) with possibly unknown transition probabilities, which characterizes how human cognitive states, such as human trust and fatigue, stochastically change with the robot performance. Under the unknown MDP models, an algorithm is developed to learn the model and obtain an optimal task allocation policy that minimizes the expected average cost for each task cycle and maximizes the probability of satisfying linear temporal logic constraints. Moreover, this paper shows that the difference between the optimal policy based on the learned model and that based on the underlying ground truth model can be bounded by arbitrarily small constant and large confidence level with sufficient samples. The case study of an assembly process demonstrates the effectiveness and benefits of our proposed learning based human robot collaboration.