Hadron Vision System
An embedded stereo 3D vision system intended for robots. It integrates an NVIDIA Jetson processor with Mirsee’s custom stereo camera and a custom motherboard to deliver “high performance and energy efficiency.
Optics & Image Sensors
Stereo camera: two cameras observe the same scene from different viewpoints; the parallax (apparent pixel shift) between the two images encodes depth.
Global shutter: a sensor that exposes all pixels simultaneously, eliminating rolling-shutter geometric distortions during motion
Hardware trigger: an electrical signal that synchronizes exposure start on both sensors to within microseconds. Why it matters: stereo assumes images are of the same scene; desync corrupts disparity
Baseline (B): distance between camera centers; larger B improves depth resolution at distance but increases minimum range and physical size.
Formula: Disparity (d): pixel offset between corresponding points in left/right images after rectification.
Z = fB/d
Mechanical Rig
A front plate with datum pins for sub-100 µm rigidity across temperature to hold sensor modules within intrinsic/extrinsic tolerances.
- Intrinsic calibration: estimates each camera’s focal lengths, principal point, distortion.
- Extrinsic calibration: estimates relative rotation & translation
Advanced Mobile Platform
Drivetrain
Purpose: a high-maneuverability base for humanoids that can strafe (translate sideways), rotate about its vertical axis in place, and carry >300 kg payload, built around mecanum wheels with integrated suspension.
Holonomic here means the platform has three independent controllable DOFs (degrees of freedom) in the plane Vx, Vy, wx (yaw angular velocity).
Why 45° rollers? It equalizes longitudinal/lateral coupling, giving symmetric performance envelopes and simpler control Jacobians (the matrix above becomes well-conditioned for a rectangular footprint).
Actuators (likely): BLDC (brushless DC) hub-coupled motors with high-ratio planetary gearboxes and encoders. Reasoning: to achieve >300 kg payload with acceleration margins and climb over thresholds, each wheel needs high stall torque
Braking: For a humanoid + payload on a slope, you need fail-safe brakes (spring-applied, electrically released) on each wheel or motor to handle power loss—especially because mecanum wheels have limited passive resistance sideways due to rollers.
Suspension
A compliant independent suspension (e.g., trailing arm with torsion bush or compact coil-over) plus a stiff but not rigid chassis improves traction uniformity and odometry consistency on imperfect floors. Literature shows mecanum bases are sensitive to unevenness; suspension mitigates this.
Power system
To run four geared BLDCs at industrial duty cycles a high-voltage battery pack (e.g., 48–60 V nominal) is typical; higher voltage lowers current for the same power.
Thermal: Wheel-motor thermal load during prolonged lateral motion can be higher (rollers dissipate energy due to micro-slip). Expect heat-sinking at motors, conductive paths to the chassis
Control Stack
Low-level control: Each wheel uses Field oriented control with current/velocity loops, reporting wheel speed, phase currents, and temperatures.
Mid-level control: A whole-body velocity controller maps a desired chassis twist [Vx,Vy,wz] to wheel speeds via the inverse kinematics with saturation management
Odometry: Integrates wheel encoder data through the forward kinematics to estimate pose. For mecanum, odometry is notoriously drift-prone under lateral commands because roller slip and unequal load bias the model; therefore the AMP likely fuses IMU (gyroscope for yaw) and LiDAR sensors
Forward kinematics calculates the end-effector’s position from the joint angles, while inverse kinematics calculates the joint angles needed to reach a specific end-effector position
Safety supervision: Watchdogs, limit zones, E-stop that cuts power and applies brakes, and tilt detection to avoid tipping under high lateral acceleration with a tall humanoid.
Haptic & Tactile Feedback
Tactile sensing at the fingers (local contact):
- Fingertip taxel arrays (pressure-sensing pixels), typically capacitive or piezoresistive, to measure normal force distribution and contact location.
- High-bandwidth slip sensors (tiny accelerometers or microphones near the skin) to detect micro-vibration that signals impending slip.
- Proximal phalanx strips for side contacts during power grasps.
Force/torque sensing in the hand and wrist (global interaction):
- Six-axis force/torque unit at the wrist to capture the net interaction wrench for safe manipulation around equipment.
- Tendon tension sensing via inline miniature load cells, or pressure sensing in hydrostatic actuators to estimate joint torques.
- Joint angle encoders to convert forces into fingertip force estimates with a hand model.
Modular Battery System
Motor Controllers
System Management Controller
Hand
Underactuated fingers: a few tendons drive multiple joints via compliant differentials for adaptive, enveloping grasps.
Proximal joint modules house pulleys/differentials;
Likely hydrostatic “M-Drive” mini-actuators in palm/forearm (or small electric motors on spools), pulling UHMWPE/Kevlar tendons.
MCP (knuckle) often directly actuated; PIP/DIP coupled by ratio for natural finger curl.