MoRL: Reinforced Reasoning for Unified Motion Understanding and Generation

TL;DR

MoRL tackles unified motion understanding and generation by training a multimodal LLM with task-specific reinforcement learning rewards and a test-time Chain-of-Motion decoder. It introduces two large-scale CoT datasets, MoUnd-CoT-140K and MoGen-CoT-140K, and a VQ-VAE–based motion tokenizer to enable discrete motion tokens for efficient, language-guided generation, with a backbone initialized from $Qwen3-4B-Instruct$ and augmented by LoRA adapters. The learning pipeline combines a cold-start supervised fine-tuning phase on synthetic CoT data with RLVR to optimize semantic alignment, reasoning coherence, physical plausibility, and text–motion consistency, followed by CoM inference to plan and reflect during generation and understanding. Experiments on HumanML3D and KIT-ML show MoRL achieves significant gains over state-of-the-art baselines across both motion understanding and generation, demonstrating improved long-horizon planning, semantic grounding, and perceptual realism. These results suggest MoRL as a practical foundation for robust motion-language reasoning in robotics, animation, and interactive systems, albeit with trade-offs in inference cost and domain-general reward design.

Abstract

Human motion understanding and generation are crucial for vision and robotics but remain limited in reasoning capability and test-time planning. We propose MoRL, a unified multimodal motion model trained with supervised fine-tuning and reinforcement learning with verifiable rewards. Our task-specific reward design combines semantic alignment and reasoning coherence for understanding with physical plausibility and text-motion consistency for generation, improving both logical reasoning and perceptual realism. To further enhance inference, we introduce Chain-of-Motion (CoM), a test-time reasoning method that enables step-by-step planning and reflection. We also construct two large-scale CoT datasets, MoUnd-CoT-140K and MoGen-CoT-140K, to align motion sequences with reasoning traces and action descriptions. Experiments on HumanML3D and KIT-ML show that MoRL achieves significant gains over state-of-the-art baselines. Code: https://github.com/AIGeeksGroup/MoRL. Website: https://aigeeksgroup.github.io/MoRL.

Figures (4)

• Figure 1: Visualization comparisons with MotionLLM. In the backflip example, MotionLLM fails to maintain a coherent takeoff-rotation-landing trajectory, resulting in unstable body orientation, while MoRL completes a physically plausible flip. In the Wack-style dance, MotionLLM shows inconsistent rotation direction and fragmented poses, whereas MoRL preserves continuous left-to-right rotation and stylistic coherence.
• Figure 2: Motion CoT data engine. Build based on MotionHubV2 dataset ling2024motionllama, one branch (MoUnd-CoT) uses motion sequences and captions with Gemini to construct reasoning chains for understanding, while the other (MoGen-CoT) builds reasoning chains for generation.
• Figure 3: Overview of MoRL. Our framework unifies motion understanding and generation under a reinforcement learning paradigm. Motion and text inputs are tokenized into a shared representation space. A hierarchical post-training pipeline first applies SFT on large-scale synthetic CoT datasets to align motion sequences with reasoning traces and concise descriptions, then employs reinforcement learning with verifiable rewards (RLVR) to refine outputs, enhancing semantic alignment, reasoning coherence, physical plausibility, and text–motion consistency. At inference, the Chain-of-Motion (CoM) decoding strategy enables step-by-step reasoning and reflection, improving both motion understanding and perceptually realistic motion generation.
• Figure 4: Results of user study.