Bridging the Resource Gap: Deploying Advanced Imitation Learning Models onto Affordable Embedded Platforms

Haizhou Ge,Ruixiang Wang,Zhu-ang Xu,Hongrui Zhu,Ruichen Deng,Yuhang Dong,Zeyu Pang,Guyue Zhou,Junyu Zhang,Lu Shi

2024-11-18

Abstract:Advanced imitation learning with structures like the transformer is increasingly demonstrating its advantages in robotics. However, deploying these large-scale models on embedded platforms remains a major challenge. In this paper, we propose a pipeline that facilitates the migration of advanced imitation learning algorithms to edge devices. The process is achieved via an efficient model compression method and a practical asynchronous parallel method Temporal Ensemble with Dropped Actions (TEDA) that enhances the smoothness of operations. To show the efficiency of the proposed pipeline, large-scale imitation learning models are trained on a server and deployed on an edge device to complete various manipulation tasks.

Machine Learning,Robotics

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the challenge of deploying advanced imitation learning algorithms on resource - limited embedded platforms. Specifically, although advanced imitation learning models based on structures such as Transformer have shown significant advantages in robotics, these large - scale models have extremely high requirements for computing resources and are difficult to be directly deployed on low - power embedded systems such as autonomous drones and mobile robots. To solve this problem, the author proposes a pipeline to promote the migration of these advanced imitation learning algorithms to edge devices through an efficient model compression method and a practical asynchronous parallel method - Temporal Ensemble with Dropped Actions (TEDA). This method not only improves the response speed of the model but also ensures that it can make real - time decisions in a dynamic environment. ### Main problem summary: 1. **Resource limitation**: Traditional large - scale imitation learning models require expensive and resource - rich computing platforms, which makes them difficult to be applied in resource - limited environments. 2. **Deployment efficiency**: Existing solutions (such as cloud - based inference) have disadvantages such as network latency, privacy issues, and insufficient reliability. 3. **Model adaptability**: A method that can effectively utilize limited computing resources is required to achieve efficient execution of complex tasks. By introducing techniques such as Input Shape Unification (ISU) and Symmetric Quantization (SQ), the author has successfully reduced the memory footprint and computing requirements of the model, thereby achieving the goal of efficiently deploying advanced imitation learning models on low - cost embedded platforms. In addition, the TEDA method further improves the smoothness and continuity of operations and solves the problem of action jitter caused by low inference frequency.

Bridging the Resource Gap: Deploying Advanced Imitation Learning Models onto Affordable Embedded Platforms

Extendable Multi-Device Collaborative Pipeline Parallel Inference in the Edge-Cloud Scenario

Robot Simulation and Reinforcement Learning Training Platform Based on Distributed Architecture.

EASTER: Learning to Split Transformers at the Edge Robustly

ED-ViT: Splitting Vision Transformer for Distributed Inference on Edge Devices

Towards Efficient Edge Learning for Large Models in Heterogeneous Resource-limited Environments.

TinyFormer: Efficient Transformer Design and Deployment on Tiny Devices

Implementation of Big AI Models for Wireless Networks with Collaborative Edge Computing

Efficient Deployment of Transformer Models in Analog In-Memory Computing Hardware

JUICER: Data-Efficient Imitation Learning for Robotic Assembly

Transformer-based deep imitation learning for dual-arm robot manipulation

Cloud-Based Hierarchical Imitation Learning for Scalable Transfer of Construction Skills from Human Workers to Assisting Robots

UAV-assisted task offloading system using dung beetle optimization algorithm & deep reinforcement learning

Optimizing the Deployment of Tiny Transformers on Low-Power MCUs

ARCADE: Scalable Demonstration Collection and Generation via Augmented Reality for Imitation Learning

Data Scaling Laws in Imitation Learning for Robotic Manipulation

MCUFormer: Deploying Vision Tranformers on Microcontrollers with Limited Memory.

LAMBO: Large AI Model Empowered Edge Intelligence

A Digital Twin Empowered Lightweight Model Sharing Scheme for Multi-Robot Systems

RISE: 3D Perception Makes Real-World Robot Imitation Simple and Effective

Generalized Robot Learning Framework