Humanoid Safety Protocol

Safety validation module for humanoid robots implementing ISO/TS 15066 compliant contact force limits, balance monitoring, and THSP validation.

Overview

The Humanoid Safety Protocol provides a comprehensive safety layer for humanoid robot operations, designed to work alongside ROS2 and simulation environments like NVIDIA Isaac Lab.

Key Features

• ISO/TS 15066 Compliance: Contact force limits based on biomechanical studies
• Balance Monitoring: ZMP/CoM tracking with fall detection
• THSP Validation: Four-gate safety validation for all actions
• Robot Presets: Pre-configured profiles for Tesla Optimus, Boston Dynamics Atlas, Figure 02
• Environment Modes: Industrial, personal care, and research configurations

Quick Start

from sentinelseed.safety.humanoid import (
    HumanoidSafetyValidator,
    HumanoidAction,
    tesla_optimus,
    BodyRegion,
)

# Create validator with Optimus configuration
validator = HumanoidSafetyValidator(
    constraints=tesla_optimus(environment="industrial"),
    strict_mode=True,
    require_purpose=True,
)

# Validate a manipulation action
action = HumanoidAction(
    joint_velocities={"left_elbow_pitch": 1.5},
    expected_contact_force=30.0,
    contact_region=BodyRegion.HAND_BACK,
    purpose="Pick up part from conveyor",
    is_collaborative=True,
)

result = validator.validate(action)

if result.is_safe:
    print("Action approved")
else:
    print(f"Blocked: {result.reasoning}")
    for violation in result.violations:
        print(f"  - {violation['description']}")

ISO/TS 15066 Contact Force Limits

The body model implements biomechanical contact force limits from the University of Mainz study, covering 29 body regions:

from sentinelseed.safety.humanoid import HumanBodyModel, BodyRegion

body_model = HumanBodyModel(safety_factor=0.9)

# Get safe force for a body region
safe_force = body_model.get_safe_force(
    region=BodyRegion.NECK_FRONT,
    contact_type="quasi_static"
)

# Check if a force is safe
is_safe = body_model.is_force_safe(
    region=BodyRegion.HAND_PALM,
    force=150.0,
    contact_type="transient"
)

Body Regions and Force Limits

Robot Presets

Tesla Optimus Gen 2

from sentinelseed.safety.humanoid import tesla_optimus

optimus = tesla_optimus(
    environment="industrial",
    include_hands=True,  # 22 DOF per hand
)

• Height: 1.73m, Weight: 70 kg
• Body DOF: 28 (+ 22 per hand optional)
• Max walking speed: 2.2 m/s (research mode)

Boston Dynamics Atlas

from sentinelseed.safety.humanoid import boston_dynamics_atlas

atlas = boston_dynamics_atlas(environment="research")

• Height: 1.5m, Weight: 89 kg
• DOF: 28
• Max running speed: 5.0 m/s
• Dynamic balance and acrobatics capable

Figure 02

from sentinelseed.safety.humanoid import figure_02

figure = figure_02(environment="industrial", include_hands=True)

• Height: 1.67m, Weight: 60 kg
• DOF: 28 (+ 16 per hand optional)
• 20+ hour battery life

THSP Validation Gates

Every action passes through four safety gates:

1. Truth Gate

Validates physical possibility:

• Joint positions within mechanical limits
• Velocities within actuator capabilities
• No NaN or infinite values

2. Harm Gate

Prevents injury and damage:

• Contact forces within ISO/TS 15066 limits
• End effector velocities for collaborative work
• Balance stability assessment

3. Scope Gate

Enforces operational boundaries:

• Position within safety zones
• Zone-specific speed limits
• Maximum operating height

4. Purpose Gate

Requires legitimate objectives:

• Explicit purpose statement (if required)
• Screens for harmful intent patterns
• Ensures beneficial action goals

Balance Monitoring

from sentinelseed.safety.humanoid import BalanceMonitor, IMUReading

monitor = BalanceMonitor()

# Update with sensor data
monitor.update_imu(IMUReading(
    roll=0.0, pitch=0.0, yaw=0.0,
    roll_rate=0.0, pitch_rate=0.0, yaw_rate=0.0,
    acc_x=0.0, acc_y=0.0, acc_z=9.81,
    timestamp=0.0,
))

# Assess balance
assessment = monitor.assess_balance()
print(f"Balance state: {assessment.state.value}")
print(f"Is safe: {assessment.is_safe}")

Balance States

Environment Modes

Integration

With ROS2

from sentinelseed.integrations.ros2 import SentinelSafetyNode
from sentinelseed.safety.humanoid import HumanoidSafetyValidator, tesla_optimus

humanoid_validator = HumanoidSafetyValidator(
    constraints=tesla_optimus(environment="industrial"),
)

node = SentinelSafetyNode()
node.add_validator("humanoid", humanoid_validator)

With Isaac Lab

from sentinelseed.integrations.isaac_lab import SentinelSafetyWrapper
from sentinelseed.safety.humanoid import HumanoidSafetyValidator, tesla_optimus

humanoid_validator = HumanoidSafetyValidator(
    constraints=tesla_optimus(environment="research"),
)

env = SentinelSafetyWrapper(
    env=isaac_lab_env,
    mode="clamp",  # Options: "block", "clamp", "warn", "monitor"
)

References

Standards

• ISO 10218:2025 - Safety requirements for industrial robots
• ISO 13482:2014 - Safety requirements for personal care robots
• ISO/TS 15066:2016 - Collaborative robots

Research

• PMC8850785 - Biomechanical limits for safe human-robot interaction (University of Mainz)