DEV Community

Cover image for Accelerating Large-Scale Robot Strategy Training: An Automated Closed Loop Architecture Based on Kiro, Trainium, and EKS

Accelerating Large-Scale Robot Strategy Training: An Automated Closed Loop Architecture Based on Kiro, Trainium, and EKS

Speaker: Junjie Tang @ AWS Amarathon 2025

Summary by Amazon Nova



Guidance for AI-Driven Robotics

  • Overview of objectives and benefits: integrate

Scalable Robotic

  • 1: NVIDIA Isaac Sim for physics-based

  • 2: Amazon EC2/EKS & Amazon Batch for scalable, parallel execution

  • 3: Amazon Bedrock foundation models, and agents via MCP server for AI

  • 4: Hugging Face LeRobot (LeRobot aims to provide models, datasets, and tools for real-world robotics in PyTorch. The goal is to lower the barrier to entry to robotics.)

  • 5: Outcome: parallel simulations

  • Cloud-native pipeline combining NVIDIA Isaac Sim, Amazon compute, Bedrock models, MCP agents

Significance and Impact of

  • Faster training Scalable fleets Real-time reasoning Continuous

  • Drastically reduces

  • Enables parallel

  • Supports real-time

  • Continuous

  • Outcome: iterative

Target Industries for

  • Where simulation-driven training delivers safer, faster, tailored

  • Manufacturing Automation: Safer Commissioning, Reduced

  • Warehouse & Logistics, Robotics

  • Retail & Delivery: Efficient

  • Healthcare Assistive Robotics: Safer Patient

  • Agricultural & Environmental Robotics

Delivery Agent from

  • Amazon Professional Services

  • A comprehensive agent system across the consulting cycle

Enterprise-Grade Quality and Security

  • Multiple validation layers mitigate AI hallucinations

  • Secure, customer-controlled environments

  • Human oversight at strategic checkpoints

  • Comprehensive security controls and protocols



AWS Professional Services (ProServe) agents

  • A multi-agent AI system architecture, for software development and delivery, associated with AWS Professional Services (ProServe) agents. The agents interact to create and manage software solutions. 

  • Sales Agent: The starting point, which initiates the process by feeding requirements or information into the workflow.

  • Delivery Agent: The central orchestrator that analyzes requirements, builds AI applications directly, and coordinates specialized work by delegating tasks to other agents.

  • Project Artifacts: An output generated from the initial input, likely documentation or initial plans, used by the Design Agent.

  • Design Agent: Takes "Project Artifacts" and produces a "Spec Package". It can also provide "Feedback" back to the Delivery Agent or the "Project Artifacts" step.

  • Spec Package: The output from the Design Agent, containing specifications for the build process.

  • Build Agent: Uses the "Spec Package" (guided by "Autopilot", an internal mechanism) to generate "Coding Artifacts".

  • Coding Artifacts: The generated code or application components resulting from the Build Agent's work.

  • Custom agents on AWS Transform: A separate, connected process that integrates with the main flow.

  • Security Agent: A persistent layer of the architecture, monitoring or enforcing security policies throughout the process.

  • Amazon Cloud stage/dev: Represents AWS environments (staging and development) where the resulting artifacts are deployed or managed.

  • Coding Artifacts are sent to the "dev" environment.

  • The "stage" environment appears to be an output or endpoint for the "Custom agents" process. 

  • The system uses intelligent agents to potentially automate and accelerate the software development lifecycle, improving efficiency and quality.



Team:

AWS FSI Customer Acceleration Hong Kong

AWS Amarathon Fan Club

AWS Community Builder Hong Kong

Top comments (0)