HomeTechnologyAutomation and RoboticsRedefining Robotics: High-Precision Autonomous Mobile Robots

Redefining Robotics: High-Precision Autonomous Mobile Robots

Courtesy: Lattice Semiconductors

Imagine a robot navigating a crowded factory floor, rerouting itself in real-time around equipment, humans, and unexpected obstacles — all while maintaining motion control and system stability. This isn’t a distant vision; this is the reality engineered by Agiliad in partnership with Lattice Semiconductor.

In a market full of autonomous mobile robots (AMRs) that rely on generic control stacks and prebuilt kits, this AMR stands out as a deep-tech system, purpose-built for intelligent indoor mobility. Unlike conventional AMRs that often trade performance for modularity or ease of deployment, this robot integrates a custom motion control framework based on Lattice’s Certus-NX FPGA, along with a ROS2-based advanced SLAM (Simultaneous Localization and Mapping), sensor fusion, and navigation stack running on NVIDIA Jetson Orin— all tightly orchestrated for low-latency, high-reliability operation.

This next-generation AMR is more than just mobile — it’s aware, adaptable, and engineered for deterministic control in real-world conditions. Designed for use in industrial settings, research labs, and beyond, the robot brings together embedded intelligence, energy efficiency, and full-stack integration to set a new benchmark for autonomous systems.

Key Features of the Robot: The Intelligence Behind the Robot

Advanced Localization & Mapping: RTAB-Map SLAM, a robust loop-closure-enabled algorithm, leverages both 3D lidar and camera feeds for consistent mapping even in environments with visual and spatial ambiguities.

  • 3D Obstacle Detection & Avoidance: Using a combination of 3D voxel layers and spatio-temporal layers, the robot dynamically detects and navigates around static and moving objects — maintaining safe clearance while recalculating routes on the fly.
  • Path Planning: The navigation stack uses the SMAC (Search-Based Motion Planning) planner for global routing and MPPI (Model Predictive Path Integral) for locally optimized trajectories, allowing real-time adaptation to dynamic environmental changes.
  • Precision Motion Control via FPGA: BLDC motors are governed by Lattice Certus-NX FPGAs executing custom PI (proportional integral) control loops in hardware, ensuring smooth acceleration, braking, and turning — critical for safety in confined spaces.

Sensor Fusion for Environmental Awareness :
Lidar and stereo camera data is processed on the Lattice Avant-E FPGA and fused with point cloud information to detect and differentiate humans and objects, providing real-time environmental awareness for safe and adaptive navigation.

System Architecture Breakdown Diagram

The AMR’s architecture is a layered, modular system built for reliability, scalability, and low power consumption. Jetson handles ROS2 algorithms, while the Lattice FPGAs manage motion control.

  • Robot Geometry and Integration with ROS2 : The robot’s geometry and joints are defined in a URDF model derived from mechanical CAD files. The Robot State Publisher node in ROS2 uses this URDF to publish robot structure and transform data across the ROS2 network.
  • Lattice Avant-E FPGA Based Sensor Fusion : Sensor data from lidar and stereo vision cameras is transmitted to the Avant-E FPGA over UDP. Avant-E employs OpenCV for real-time image identification and classification, fusing visual data with point cloud information to accurately detect and differentiate humans from other objects in the environment. This fused data — including human-specific classification and distance metrics — is then transmitted to the ROS2 framework running on NVIDIA Jetson. This high-fidelity sensor fusion layer ensures enhanced situational awareness, enabling the robot to make informed navigation decisions in complex, dynamic settings.
  • SLAM & Localization: Lidar provides a 3D point cloud of the environment, while the camera supplies raw image data. An RTAB-Map (Real-Time Appearance-Based Mapping) processes this information to create a 3D occupancy grid. Odometry is derived using an iterative closest point (ICP) algorithm, with loop closure performed using image data. This enables continuous optimization of the robot’s position, even in repetitive or cluttered spaces.
  • Navigation: Navigation generates cost maps by inflating areas around obstacles. These cost gradients guide planners to generate low-risk paths. SMAC provides long-range planning, while MPPI evaluates multiple trajectory options and selects the safest path.
  • ROS2 Control and Differential Drive Kinematics: ROS2 computes a command velocity (linear and angular) which is translated into individual wheel velocities using differential drive kinematics.
  • Hardware Interface: This layer ensures integration between ROS2 and the robot’s hardware. Serial communication (UART) between Jetson and Certus-NX transmits motor velocity commands in real-time.
  • Lattice Certus-NX FPGA-Based Motion Control: Lattice’s Certus-NX FPGA executes real-time motor control algorithms with high reliability and minimal latency, enabling deterministic performance, efficient power use, and improved safety under industrial loads:

PI Control Loops for velocity and torque regulation, using encoder feedback to ensure performance regardless of frictional surface conditions.

Commutation Sequencer that uses hall sensor feedback to control 3-phase BLDC motor excitation.

 

How It All Works Together: A Decision-Making Snapshot

The robot’s intelligence simulates a real-time decision-making loop:

Where am I?

The robot localizes using RTAB-Map SLAM with loop closure, updating its position based on visual and spatial cues.

Where should I go?
A user-defined goal (set via touchscreen or remote interface) is passed to the global planner, which calculates a safe, efficient route using SMAC.

How do I get there?
The MPPI planner simulates and evaluates dozens of trajectories in real-time, using critic-based scoring to dynamically adapt to the robot’s surroundings.

What if something blocks the path?
Sensor data updates the obstacle map, triggering real-time replanning. If no safe path is found, recovery behaviors are activated via behavior servers.

Component / Design Element Rationale
Differential Drive Simpler control logic and reduced energy usage compared to omni-wheels
Lidar Placement (Center) Avoids blind spots; improves loop closure and mapping accuracy
Maxon BLDC Motors High torque (>4.5 Nm) for payload handling and smooth mobility
Certus-NX FPGA Motion Control Enables deterministic control with low CPU overhead
Camera Integration Improves visual SLAM and scene understanding
Convex Caster Wheels Reduces ground friction, enhances turning in confined areas
Cooling Architecture Fans and vents maintain safe operating temperatures
Jetson as CPU Provides headroom for future GPU-based algorithm integration

Lattice FPGA Technology
Lattice’s Certus-NX and Avant-E FPGAs deliver complementary capabilities that are critical for autonomous robotic systems:

  • Low Power Consumption : Extends battery life in mobile systems
  • Real-Time Performance: Delivers responsive control loops and fast data handling
  • Flexible Architecture : Supports custom control logic and sensor interfaces

Combined with NVIDIA Jetson Orin and embedded vision tools, the result is a scalable and adaptable robotic platform.

Looking Ahead: Enabling the Future of Robotics
Agiliad’s engineering model emphasizes deep system-level thinking, rapid prototyping, and cross-domain integration, delivering a fully operational system within a compressed development timeline by leveraging low power Lattice FPGAs. This reflects Agiliad’s deep expertise in full-stack design and multidisciplinary integration across mechanical, electrical, embedded, and software.

Ralated Articles

Latest Posts