Robots are now expected to execute increasingly complex long-horizon tasks in unstructured environments. Despite the strong potential of pretrained Vision-Language Models (VLMs) in task planning, their direct application to robotic manipulation is hindered by logical reasoning deviations and inadequate geometric scene perception. This work proposes an adaptive task planning method based on video priors and dynamic scene graphs (ATP-VPDSG). It leverages the VLM to extract manipulation logic from video demonstrations, thus supplementing manipulation priors. Meanwhile, scene graphs were integrated to convert unstructured environments into structured representations with spatial topological relations, compensating for perceptual deficiencies. A dual-track feedback mechanism based on visual expectations was further incorporated to enable failure diagnosis and adaptive replanning in complex environments. Extensive long-horizon robotic manipulation experiments were conducted on the LIBERO-10 benchmark with Qwen3-VL as the core VLM. Results showed that ATP-VPDSG achieved an average task planning accuracy of 91.2% and a task execution success rate of 74.67%, outperforming the selected task planning baselines. Ablation studies verified that video priors and dynamic scene graphs exerted complementary effects on logical constraints and physical feasibility. Furthermore, a real-robot experiment on an industrial slider–rail assembly task demonstrated successful sim-to-real transfer, achieving an 82.0% success rate without task-specific fine-tuning.

Leveraging discrete sequence prediction for manipulation skill learning has demonstrated immense potential for complex long-horizon tasks. However, guaranteeing precise and robust execution under complex scenarios remains a core challenge for efficient autonomous robotic operation. This demands execution policies capable of generating precise and efficient action sequences based on perceptual feedback. To address the loss of fine-grained action details induced by action discretization and the difficulty in preserving long-horizon temporal dependencies in long-horizon tasks, this work proposed a hierarchical manipulation skill learning method based on residual finite scalar quantization (RFSQ). At the action representation level, a robust RFSQ with an adaptive linear scaling mechanism was introduced. It effectively mitigated residual decay in multi-level action quantization and realized accurate discrete reconstruction of continuous action sequences. At the action sequence generation level, a hybrid hierarchical architecture integrating Mamba2 and Transformer was constructed. Mamba2 efficiently captured temporal dependencies among actions, while Transformer precisely modeled fine-grained residual dependencies across action quantization hierarchies. A multimodal perception fusion module was further incorporated to strengthen instruction-semantic alignment on language-instructed tasks. A series of experiments were conducted on the LIBERO and Meta-World simulation environments. Experimental results demonstrated that the proposed method achieved outstanding performance in multi-task and long-horizon task learning.