Skip to Content

Embodied AI & Physical Robotics

Welcome to the isA Embodied AI & Physical Robotics Developer Center.

The isA Embodied AI platform bridges foundation model cognitive reasoning with physical reality actuation. Through the Robot Abstraction Layer (RAL), 200Hz closed-loop edge bus, and isA Agent SDK 0.6.0, developers can safely connect autonomous AI agents to physical robotic systems, bionic dexterous hands, and industrial manipulators.

1. Architectural Philosophy

Physical hardware demands microsecond-level latency guarantees, deterministic execution, and hardware emergency stop lines that fundamentally diverge from pure software API workflows.

Core Principle: Real-Time Loops at the Edge, Governance & Durability in the Cloud

  • Edge Device Tier (RAL): High-frequency closed-loop control (100 Hz ~ 200 Hz), sensor verification receipts, physical boundary limit enforcement, and emergency stops strictly reside on the physical machine locally.
  • Cloud Control Plane (isagent.io): Session lease management, durable job scheduling, cryptographic audit trails, signed OTA firmware rollout, and multi-tenant telemetry execute on Google Cloud Platform.

Cloud-Edge Collaborative Data Architecture

┌────────────────────────────────────────────────────────┐
│            Autonomous AI Agent (SDK 0.6.0)             │
│   Reasoning Loop ──► Tool Invocation (execute_skill)   │
└───────────────────────────┬────────────────────────────┘
                            │ HTTPS / MCP Tool Call
                            ▼
┌────────────────────────────────────────────────────────┐
│             GCP Platform Gateway & Governance          │
│   • Northbound Gateway: wss://isagent.io/mqtt          │
│   • user-auth: Mutual mTLS & Device Identity Tokens    │
│   • user-device: Durable Command Ledger & Idempotency  │
└───────────────────────────┬────────────────────────────┘
                            │ WSS / MQTTS (Exclusive Subscription)
                            ▼
┌────────────────────────────────────────────────────────┐
│         Edge Hardware Watchdog & Execution Daemon      │
│   • 60-Second Timed Exclusive Lease Guard              │
│   • Zero Automatic Retries Physical Barrier            │
└───────────────────────────┬────────────────────────────┘
                            │ Internal Process / CAN FD Industrial Bus
                            ▼
┌────────────────────────────────────────────────────────┐
│             Robot Abstraction Layer (RAL)              │
│   • RP Hand Adapter (Stroke: 0..10000, Current: 0..3000)│
│   • ORCA 7-DoF Industrial Arm (Radians/m, Constant Nm) │
│   • Real Sensor Closed-Loop (completion_verified)      │
└────────────────────────────────────────────────────────┘

2. Core Pillars

  1. RAL Protocol & Register Standards: Native units standard, register mapping, and zero-confusion driver abstraction.
  2. RP Hand Dexterous Control: 6 active degrees of freedom, high-frequency tactile feedback, and microcurrent tuning.
  3. Robotic Arms & Force Control: 7-axis Cartesian trajectory planning, impedance control, and constant surface compliance.
  4. Cloud-Edge Safety & Leases: 60-second exclusive leases, CAN FD disconnect handling, and physical power isolation.

3. Quickstart Code Example (Python)

from isa_agent_sdk import AgentClient
from isa_agent_sdk.embodied import RPHandDevice, GraspMode
 
client = AgentClient.from_env()
hand = RPHandDevice.connect("can0", baudrate=1_000_000)
 
@client.tool(name="soft_grasp")
async def soft_grasp(target_force_g: int = 150):
    """Drive bionic dexterous hand with closed-loop tactile force feedback via RAL"""
    result = await hand.execute_grasp(
        mode=GraspMode.ADAPTIVE_ENVELOP,
        max_current_limit=800,
        target_force_grams=target_force_g,
        watchdog_lease_seconds=30
    )
    return {
        "status": "success" if result.completion_verified else "safety_hold",
        "applied_force_g": result.measured_force_grams,
        "encoder_positions": result.joint_positions
    }
 
agent = client.create_agent(
    model="gemini-2.5-pro",
    tools=[soft_grasp],
    system_prompt="You are a high-precision embodied robotics agent executing delicate pick-and-place tasks."
)