RoboMT: Human-Like Compliance Control for Assembly Via a Bilateral Robotic Teleoperation and Hybrid Mamba-Transformer Framework
RUNDONG WANG, Yanchun Cheng, Qilong Yuan, Alok Prakash, Francis TAY, Marcelo H Ang Jr
AI summary
Problem
Traditional compliance control methods struggle with modeling inaccuracies, sensor noise, and inefficient processing of long-sequence interaction data, while comprehensive real-world force-motion datasets remain scarce.
Approach
The authors developed a bilateral teleoperation platform to collect human-guided force and motion data, then trained a hybrid Mamba-Transformer framework with adaptive action chunking to predict compliant robot adjustments from contact forces.
Key results
- Novel bilateral teleoperation platform and long-sequence force-motion dataset
- Hybrid Mamba-Transformer architecture enabling near-linear sequence modeling
- Superior success rates (62–98%) and stable force regulation (~2.5 N) over baselines
- Real-time inference at 55 ms per batch with minimal force overshoot
Why it matters
Provides a scalable, data-driven pathway for precise and adaptable robotic assembly, benefiting automation engineers and embodied AI researchers.
Abstract
Robotic compliance control is critical for delicate tasks such as electronic connector assembly, where precise force regulation and adaptability are paramount. However, traditional methods often struggle with modeling inaccuracies and sensor noise. Inspired by human adaptability in complex assembly operations, we present RoboMT, a novel framework that in- tegrates a Mamba algorithm with a Transformer architecture to achieve human-like compliance control. By leveraging a bilateral teleoperation platform, we collect extensive real-time force/torque and motion data to form a comprehensive dataset for training. Furthermore, RoboMT incorporates an Adaptive Action Chunk module and a Temporal Fusion module to ensure smooth and robust action prediction. Experimental results across four electronic assembly tasks show that RoboMT achieves superior success rates (62–98%) over baselines (29–98%), while maintaining stable force regulation around 2.5N, closely resem- bling human performance. During task transitions, RoboMT quickly stabilizes at 5N with minimal overshoot, avoiding the large force spikes (over 24N) seen in baselines. Additionally, RoboMT maintains an average inference speed of 55 ms per batch, balancing real-time responsiveness and control robustness. Overall, RoboMT presents a compelling pathway toward error- minimized, human-level compliance control, and generalization for real-world robotic assembly, setting a new benchmark for precision, adaptability, and robustness in robotic assembly.