Hierarchical Proportion Models for Motion Generation via Integration of Motion Primitives

TL;DR

Imitation learning for long-horizon robotic tasks typically requires large datasets; this work introduces a two-layer hierarchical IL that decomposes tasks into motion primitives and combines them with proportions to synthesize motion. The upper layer provides long-horizon planning while the lower layer yields primitive outputs, with three proportion strategies: learning-based, sampling-based (MC-MPC-inspired), and playback-based, which differ in how the proportions are set and whether the upper layer is trainable. Actions are produced as a weighted sum of primitives, $u = \sum_i p_i a_i$, with $p_i$ constrained by a softmax, and sampling methods use a cross-entropy weighting to select the best candidates. Real-robot pick-and-place experiments validate the approach, showing that sampling-based and playback-based variants yield higher stability and adaptability than the baseline, enabling generation of complex motions not seen in the primitive set and enabling reuse of lower-layer primitives to reduce training cost.

Abstract

Imitation learning (IL) enables robots to acquire human-like motion skills from demonstrations, but it still requires extensive high-quality data and retraining to handle complex or long-horizon tasks. To improve data efficiency and adaptability, this study proposes a hierarchical IL framework that integrates motion primitives with proportion-based motion synthesis. The proposed method employs a two-layer architecture, where the upper layer performs long-term planning, while a set of lower-layer models learn individual motion primitives, which are combined according to specific proportions. Three model variants are introduced to explore different trade-offs between learning flexibility, computational cost, and adaptability: a learning-based proportion model, a sampling-based proportion model, and a playback-based proportion model, which differ in how the proportions are determined and whether the upper layer is trainable. Through real-robot pick-and-place experiments, the proposed models successfully generated complex motions not included in the primitive set. The sampling-based and playback-based proportion models achieved more stable and adaptable motion generation than the standard hierarchical model, demonstrating the effectiveness of proportion-based motion integration for practical robot learning.

Figures (8)

• Figure 1: Structure of hierarchical model.
• Figure 2: Algorithm of MC-MPC.
• Figure 3: Comparison between different data separation methods.
• Figure 4: Structure of the learning-based proportion model.
• Figure 5: Algorithm of proposed sampling-based proportion model.