Autonomous Mobile Robots Design for E-Commerce Fulfillment

With increasing operational demands and customer expectations in e-commerce centers, there is a compelling need for technologies that can scale to meet these evolving challenges. One technology showing remarkable efficiency is Autonomous Mobile Robots (AMRs). Deployed to streamline workflows and significantly reduce operational costs, AMRs are rapidly becoming an integral part of e-commerce fulfillment centers across the globe.

According to a report by ABI Research¹, over four million commercial AMRs will be operational in over 50,000 warehouses around the world by 2025. This paper will discuss the functionalities and mechanisms which allow AMRs to operate efficiently, design considerations for AMRs, and the Qualcomm® Robotics RB5 Development Kit for robotic IoT applications. 

 

Basics of AMR Operation

AMRs are a class of robots designed to move autonomously within environments without the need for pre-defined paths or external guidance systems. In warehouses, AMRs are often used to execute various receiving, picking, sorting, packaging, and transporting tasks. 

AMRs exhibit the following attributes:

Autonomous Operation

AMRs can function without continuous human input, making decisions based on predefined algorithms and real-time data. Additionally, AMRs work around the clock, increasing productivity and reducing errors associated with human fatigue.

Mobility Across Varied Terrains within a Warehouse

Unlike traditional Automated Guided Vehicles (AGVs) that rely on tracks, AMRs can navigate freely within warehouse environments. They can avoid obstacles, reroute, and adapt to real-time changes. This flexibility ensures operational continuity, even when there are modifications to the warehouse layout, obstacles, or traffic congestion. 

Enhancement and Assistance of Human-driven Processes

AMRs are designed to complement or enhance human input with arms, shelves, and other tools to pick, pack, or transport items. Collaboration between AMRs and humans results in increased productivity. While AMRs handle repetitive physical tasks, humans can focus on other activities, increasing operational efficiency.

 

AMR Functionality

AMRs comprise an array of features and rely heavily on heterogeneous computing. The basis of their operations is understanding their immediate environment and making intelligent navigation decisions. AMR functionality can be categorized into four main capabilities: sensing, processing (thinking), actuating (acting), and communication. 

Figure 1. Typical Autonomous Mobile Robot (AMR) used in e-commerce fulfillment

 

Seeing Surroundings in 3D

To interact with the world around them, AMRs must first understand it. This begins by perceiving their surroundings in 3D. 

Key AMR components for 3D viewing include the following:

• Structured Light Camera: Decodes a scene by projecting a pattern of pixels and measuring the distortion when these pixels bounce back. The resulting data can provide depth information.
• Time-of-Flight Camera: Utilizes the principle of measuring the time taken for a light signal to travel to an object, reflect, and return to the sensor. This process allows the device to determine distances.
• Stereo Cameras: Capture images with two or more cameras placed at different positions, and when combined, these systems can generate depth maps from parallax.
• LIDAR: Sends laser beams and interprets the reflected signals to gauge distances.
• SONAR: Emits sound pulses and listens for their echoes to calculate distances.

Creating a 3D Map of its Surroundings

After receiving data from the sensors and cameras, AMRs use Simultaneous Localization and Mapping (SLAM) to generate a dynamic 3D map of their surroundings as they navigate through it. This dynamic 3D representation evolves as the AMR moves and encounters new elements.

Localization

An AMR needs to identify its position within the generated map. This is achieved by combining motion data from the camera feeds with inertial data from onboard sensors and a wheel encoder to estimate motion and improve its localization accuracy.

There are two approaches to SLAM:

• Visual SLAM: Incorporates a camera and an Inertial Measurement Unit (IMU) to gauge position and orientation.
• LIDAR SLAM: Uses a laser sensor in tandem with an IMU. LIDAR technology is more accurate for depth estimation.

To improve localization, private 5G networks in warehouses and fulfillment centers can bolster SLAM techniques, adding another layer of positioning accuracy.

Navigation

Once localized, AMRs set about their primary task: navigation. This process involves:

• Scene Understanding: Providing a spatial and semantic understanding of the environment using depth sensors coupled with machine learning algorithms.
• Path Planning: Finding the optimal path through the environment and meeting high-level goals while avoiding obstacles.
• Real-time Control: Implementing a motion plan by translating desired speed and direction into motor commands. 
• Motion Estimation: Constantly re-evaluating its position on the map as the AMR moves. Any discrepancies lead to dynamic path updates. 

Navigation involves adapting to environmental changes like people, shelves, and walls. AMRs rely on LIDAR to detect changes and use machine learning to refine navigation goals. They also take advantage of Indoor Precise Positioning (IPP), using 5G Transmission Points/Reception Points (TRP) to plot a grid they can use for centimeter-level accuracy on the x-, y-, and z-axes. 

Object Recognition and Obstacle Avoidance

An integral part of an AMR's autonomy is its ability to identify objects and avoid obstacles. Such tasks are highly reliant on:

• Computer Vision: Enables AMRs to interpret visual data from the world.
• Artificial Intelligence: Enhances the AMRs’ understanding of objects and learn from previous interactions, improving overall efficiency.

AMRs execute the above functions on-device for optimal performance and real-time decisions, avoiding the latency of cloud-based communications.

 

AMR Design Considerations for E-Commerce Fulfillment

The unique requirements of e-commerce fulfillment centers, such as high throughput, variable package sizes, and real-time order changes, impose specific design constraints on AMRs. 

The following are key design considerations:

• Payload and Modularity: AMRs should have a scalable payload capacity, considering the varying sizes of items. Modular designs can help to customize robot capabilities based on operational requirements.
• Energy Efficiency: AMRs require long battery life and quick charging or battery-swapping capabilities to achieve extended operational hours. 
• Crowded/Space-Constrained Navigation: AMRs must feature collision avoidance, path optimization, and efficient rerouting capabilities due to the crowded nature of fulfillment centers.

 

Qualcomm Robotics RB5 Development Kit: Powerful Robotics for AMRs

The Qualcomm Robotics RB5 Development Kit provides a robust platform for developing diverse robotic applications, encompassing AMRs, drones, and more. Built around the Qualcomm® QRB5165 processor, this development kit delivers high-speed wireless connectivity and high-accuracy AI/ML inferencing technology to accelerate the development of power-efficient, high-computing robots, and drones. It provides multiple software options, including support for embedded Linux and Ubuntu. 

The Qualcomm Robotics RB5 Development Kit can be used as the foundation for developing AMRs (such as the one shown in Figure 2), drones, and many other robotics applications spanning manufacturing, mapping, logistics, delivery, health care, retail, and more.

 

Figure 2. An AMR developed using the Qualcomm Robotics RB5 Development Kit  

 

Key Specifications

The Qualcomm Robotics RB5 Development Kit and QRB5165 processor include the following key components:

• Qualcomm® Kryo™ 585 CPU built on Arm® Cortex® technology
• Qualcomm® Adreno™ 650 GPU
• Qualcomm® Hexagon™ DSP with quad Hexagon Vector eXtensions (HVX) processor for vision processing and machine learning
• Qualcomm Spectra™ 480 image processing engine
• Adreno VPU 665 for high-quality, ultra HD video encode and decode
• Adreno DPU 995 for on-device and external ultra HD display support
• Low-power audio subsystem combined with the Qualcomm Aqstic™ Audio Technologies WCD9380/WCD9385 audio codec for low-power voice processing and audiophile quality audio playback
• Qualcomm® Sensing Hub for contextual awareness and always-on sensor support
• Qualcomm® Secure Processing Unit (SPU240) for advanced security-focused use cases
• Qualcomm® Neural Processing Unit (NPU230) for high-performance machine learning use cases
• External 802.11ax, 2 × 2 MIMO, and Bluetooth® 5.1
• Quad-channel package-on-package (PoP) high-speed LPDDR5 SDRAM

 

Figure 3. QRB5165 processor high-level block diagram and outline drawing 

 

Memory

The EBI0 and EBI1 ports are dedicated to the PoP LPDDR5 SDRAM memory that is attached to the top of the QRB5165 processor. 

Audio

The QRB5165 processor supports the WCD9380/WCD9385 audio codec IC to provide the system’s audio functions. QRB5165 audio-related interface options with the WCD include the following:

• Digital microphone PDM interface
• SoundWire (SWR) interface
• SLIMbus interface
• I2S interfaces
• PCM/TDM interfaces
• I2C/I3C interface

Display

The QRB5165 processor supports up to two D-PHY or C-PHY displays.

DMB

The QRB5165 processor supports an external DMB solution using SPI or SD interface options.

 

Conclusion

The future of e-commerce fulfillment is inseparable from the advancements in AMR technology. The increasing capabilities of these robots, combined with advancements in AI and ML, should continue to drive their adoption and evolution into the future. Ensuring a deep understanding of AMR functionality is essential for maximizing their potential in modern warehouses. The Qualcomm Robotics RB5 Development Kit provides a robust platform for developing high-performance and power-efficient AMRs for warehouse applications. 

For more information, see the Qualcomm Robotics RB5 developer portal.

All images courtesy of Qualcomm Technologies, Inc.

¹50,000 Warehouses to Use Robots by 2025 as Barriers to Entry Fall and AI Innovation Accelerates

Note: Qualcomm branded products are products of Qualcomm Technologies, Inc. and/or its subsidiaries. Arm and Cortex are registered trademarks of Arm Limited (or its subsidiaries or affiliates) in the US and/or elsewhere.