Skip to main content

Case Study: Amazon's Deployment of Agility Digit Robots

Summary​

In October 2023, Amazon announced a pilot program to test Agility Robotics' Digit humanoid robot in its fulfillment centers. By 2024, the program expanded to multiple facilities, focusing on tote handling and recycling tasks. This represents one of the largest commercial deployments of bipedal robots in logistics.

Background​

Agility Robotics and Digit​

Agility Robotics, a spin-off from Oregon State University's ATRIAS project, developed Digit specifically for logistics and warehouse applications.

Digit Specifications:

  • Height: 5'9" (175 cm)
  • Weight: 141 lbs (64 kg)
  • Payload: 35 lbs (16 kg)
  • Walking speed: Up to 1.5 m/s
  • Battery life: 8+ hours
  • Manipulation: 4-DOF arms with parallel gripper

Amazon's Robotics Strategy​

Amazon has invested heavily in warehouse automation:

  • 750,000+ robots deployed (as of 2024)
  • Acquisition of Kiva Systems (2012)
  • Development of Proteus autonomous mobile robot
  • Investment in Agility Robotics ($150M+)

Technical Implementation​

Primary Use Case: Tote Recycling​

Digit robots handle empty plastic totes in the fulfillment workflow:

  1. Detection: Identify empty totes on conveyors
  2. Pick: Grasp totes from conveyor or storage
  3. Transport: Walk to destination area
  4. Place: Stack or organize totes for reuse

System Architecture​

class DigitWarehouseTask:
    """
    Digit integration with Amazon fulfillment systems
    """
    def __init__(self, robot_id, facility_zone):
        self.robot_id = robot_id
        self.zone = facility_zone
        self.wms_client = WarehouseManagementClient()

    async def execute_tote_task(self, task):
        # Receive task from WMS
        pickup = await self.wms_client.get_pickup_location(task)
        dropoff = await self.wms_client.get_dropoff_location(task)

        # Navigate to pickup
        await self.navigate_to(pickup.location)

        # Perception and grasp
        tote_pose = self.perceive_tote()
        await self.pick_tote(tote_pose)

        # Transport
        await self.navigate_to(dropoff.location)

        # Place
        await self.place_tote(dropoff.slot)

        # Report completion
        await self.wms_client.report_complete(task)

Integration Challenges​

ChallengeSolution
Floor variabilityRobust locomotion control with terrain adaptation
Dynamic obstaclesReal-time path replanning, human detection
Tote variationsAdaptive grasping with force sensing
System integrationREST APIs connecting to Amazon WMS

Locomotion on Warehouse Floors​

Challenges Specific to Warehouses​

  • Surface transitions: Concrete to anti-fatigue mats
  • Debris: Small items on floor
  • Slopes: Loading dock ramps
  • Wet areas: Near wash stations

Control Approach​

Digit uses a hierarchical control architecture:

  1. High-Level Planner: Path planning with A* or RRT
  2. Mid-Level Footstep Planner: Foot placement optimization
  3. Low-Level Controller: Real-time balance and locomotion
# Simplified locomotion hierarchy
class DigitLocomotion:
    def navigate_to(self, goal):
        # High-level path
        path = self.path_planner.plan(self.position, goal)

        for waypoint in path:
            # Generate footstep plan
            footsteps = self.footstep_planner.plan(
                current=self.foot_positions,
                target=waypoint,
                terrain=self.terrain_map
            )

            # Execute with balance control
            for step in footsteps:
                self.step_controller.execute(step)
                self.balance_controller.maintain_stability()

Outcomes​

Pilot Program Results​

  • Successful handling of thousands of totes daily
  • Reduced ergonomic strain on human workers
  • Integration with existing fulfillment workflows
  • Data collection for continued improvement

Scalability Considerations​

  • Agility building dedicated manufacturing facility
  • Target: 10,000+ units annual production capacity
  • Multi-facility deployment planned

Safety Architecture​

Human-Robot Collaboration​

Digit operates in shared spaces with human workers:

  1. Detection: 360° awareness of human presence
  2. Speed Reduction: Automatic slowdown near humans
  3. Stop: Emergency stop capability
  4. Communication: Visual indicators of robot state

Compliance​

  • OSHA guidelines for material handling
  • ANSI/RIA R15.06 industrial robot safety
  • Amazon internal safety standards

Economic Analysis​

Cost-Benefit Considerations​

FactorImpact
Unit Cost~$250,000 per Digit robot
Labor SavingsRepetitive task automation
Injury ReductionErgonomic benefit quantification
FlexibilityAdaptable to multiple tasks
ScalabilityIncremental deployment possible

ROI Factors​

  • Task throughput comparison
  • Maintenance and support costs
  • Training and integration expenses
  • Productivity gains from human workers

Discussion Questions​

  1. Why did Amazon choose bipedal robots over wheeled alternatives for this application?
  2. How does the warehouse environment differ from factory settings for robot deployment?
  3. What are the implications of large retailers driving humanoid robot development?
  4. How should safety standards evolve for human-robot collaboration in logistics?
  • Module 06: Motion Planning - Path planning in dynamic environments
  • Module 08: Locomotion - Bipedal walking control
  • Module 10: Simulation to Real - Deployment in unstructured environments

External References​

Current as of: December 2024