Towards Closed-Loop Embodied Empathy Evolution: Probing LLM-Centric Lifelong Empathic Motion Generation in Unseen Scenarios

2025/12/22 by Wang, Jiawen, Chen, Jingjing Wang Tianyang, Zhang, Min +1
#Artificial Intelligence (cs.AI) #FOS: Computer and information sciences

paper · doi:10.48550/arxiv.2512.19551

Abstract

In the literature, existing human-centric emotional motion generation methods primarily focus on boosting performance within a single scale-fixed dataset, largely neglecting the flexible and scale-increasing motion scenarios (e.g., sports, dance), whereas effectively learning these newly emerging scenarios can significantly enhance the model's real-world generalization ability. Inspired by this, this paper proposes a new LLM-Centric Lifelong Empathic Motion Generation (L²-EMG) task, which aims to equip LLMs with the capability to continually acquire emotional motion generation knowledge across different unseen scenarios, potentially contributing to building a closed-loop and self-evolving embodied agent equipped with both empathy and intelligence. Further, this paper poses two key challenges in the L²-EMG task, i.e., the emotion decoupling challenge and the scenario adapting challenge. To this end, this paper proposes an Emotion-Transferable and Scenario-Adapted Mixture of Experts (ES-MoE) approach which designs a causal-guided emotion decoupling block and a scenario-adapted expert constructing block to address the two challenges, respectively. Especially, this paper constructs multiple L²-EMG datasets to validate the effectiveness of the ES-MoE approach. Extensive evaluations show that ES-MoE outperforms advanced baselines.

Citations

• ZeroHSI: Zero-Shot 4D Human-Scene Interaction by Video Generation
• Learning from Massive Human Videos for Universal Humanoid Pose Control
• UniMuMo: Unified Text, Music and Motion Generation
• Autonomous Character-Scene Interaction Synthesis from Text Instruction
• Scalable Language Model with Generalized Continual Learning
• Boosting Continual Learning of Vision-Language Models via Mixture-of-Experts Adapters
• Higher Layers Need More LoRA Experts
• MoMask: Generative Masked Modeling of 3D Human Motions
• Motion-X: A Large-scale 3D Expressive Whole-body Human Motion Dataset
• MotionGPT: Human Motion as a Foreign Language
• ReMoDiffuse: Retrieval-Augmented Motion Diffusion Model
• Executing your Commands via Motion Diffusion in Latent Space
• EDGE: Editable Dance Generation From Music
• MotionDiffuse: Text-Driven Human Motion Generation with Diffusion Model
• MotionCLIP: Exposing Human Motion Generation to CLIP Space
• Causal Disentanglement for Semantics-Aware Intent Learning in Recommendation
• Synthesis of Compositional Animations from Textual Descriptions
• DanceFormer: Music Conditioned 3D Dance Generation with Parametric Motion Transformer
• Learning Trajectory Dependencies for Human Motion Prediction
• Selective Experience Replay for Lifelong Learning
• Neural Discrete Representation Learning
• Gradient Episodic Memory for Continual Learning
• Learning without Forgetting
• Show, Attend and Tell: Neural Image Caption Generation with Visual Attention
• Exact solutions to the nonlinear dynamics of learning in deep linear neural networks
• ChatPose: Chatting about 3D Human Pose

Related

• Emotionally Numb or Empathetic? Evaluating How LLMs Feel Using EmotionBench
• Large Emotional World Model
• EVOLVE: Emotion and Visual Output Learning via LLM Evaluation
• SYNTHEMPATHY: A Scalable Empathy Corpus Generated Using LLMs Without Any Crowdsourcing
• Enhancing Emotional Generation Capability of Large Language Models via Emotional Chain-of-Thought