What does this research mean for the field?

The Hierarchical Decision Mamba-Liquid (HDML) architecture enables rapid few-shot transfer across diverse robotic morphologies and achieves significantly higher robustness to physical perturbations compared to Decision Transformers while maintaining low inference latency. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

To develop a scalable foundation architecture for multi-embodiment continuous robotic control that resolves long-horizon reasoning and high-frequency, continuous-time motor execution without torque chatter or quadratic computational overhead.

August 23, 2026Open Access

Hierarchical Decision Mamba-Liquid Architecture for Multi-Embodiment Continuous Robotic Control

View Full Paper

Ask AI

Bookmark

To develop a scalable foundation architecture for multi-embodiment continuous robotic control that resolves long-horizon reasoning and high-frequency, continuous-time motor execution without torque chatter or quadratic computational overhead.
Designed the HDML architecture coupling an 8-layer Selective State Space Model (Mamba-3) with Rotary Position Embeddings (RoPE) for O(1) streaming inference and a Closed-Form Continuous-Time (CfC) Neural ODE motor filter.
Pre-trained the model across 11 robotic morphologies on 1,735,673 physical transitions and evaluated rapid few-shot adaptation by freezing 95.9%–97.0% (11.99M) backbone parameters while fine-tuning 376k–516k adapter parameters in under 30 seconds.
Achieved an action loss of 0.00602 on a 12-DoF Unitree A1 quadruped and 0.02579 on a 17-DoF 3D Humanoid during rapid adapter adaptation.
Attained a 16-fold higher Interquartile Mean (IQM) reward over Decision Transformers (18.88 vs. 1.17) under physical force impulses and sensor noise perturbations.
Maintained a model inference latency of 12.81 ms (78.0 Hz) and an end-to-end closed-loop control frequency of 43.2 Hz (23.1 ms cycle time including physics simulation and safety monitoring).

Discussion

No takes yet. Share an insight, caveat, or question.

Implication

Computational study demonstrates robust multi-embodiment continuous robotic control using a Mamba-Liquid architecture, indicating superior adaptability and noise resilience over Transformers.

Cite This Study

Tri Kien Ha (2026) studied this question.

synapsesocial.com/papers/6a8aae207677a34114446e61 https://doi.org/10.5281/zenodo.22020016

View Full Paper

Ask AI

Bookmark

Implication

Computational study demonstrates robust multi-embodiment continuous robotic control using a Mamba-Liquid architecture, indicating superior adaptability and noise resilience over Transformers.

Hierarchical Decision Mamba-Liquid Architecture for Multi-Embodiment Continuous Robotic Control

Key Points

Discussion

Implication

Cite This Study

Hierarchical Decision Mamba-Liquid Architecture for Multi-Embodiment Continuous Robotic Control

Key Points

Discussion

Implication

Cite This Study

Also Consider

Also Consider