Skip to content

Service Robots

Overview

Service robots operate in non-manufacturing environments to serve humans, spanning delivery, warehousing, food service, cleaning, companionship, and other segments. Unlike industrial robots, service robots must operate safely in unstructured, human-coexisting environments, placing higher demands on perception, navigation, and human-robot interaction.

Market Landscape

Service robots are the fastest-growing sector of the robotics industry:

  • Delivery/logistics: Largest commercialization track, AMR market growing rapidly
  • Cleaning: Both consumer (robot vacuums) and commercial (floor scrubbers) are mature
  • Food service/hospitality: Delivery robots have been deployed at scale in China
  • Companionship/social: Technology frontier, but difficult to commercialize

Delivery Robots

Last-Mile Delivery

Last-mile delivery robots autonomously deliver small packages/food on sidewalks or within closed campuses:

Platform Company Features Deployment Status
Nuro R3 Nuro Autonomous delivery vehicle, road driving Operating in multiple US cities
Starship Starship Technologies Sidewalk delivery, 6-wheel University campuses, communities
Serve Uber (formerly Serve Robotics) Sidewalk delivery Operating in Los Angeles
Meituan Autonomous Delivery Meituan Outdoor autonomous delivery Beijing Shunyi, etc.
Segway Delivery Robot Segway-Ninebot Indoor + elevator + outdoor Hotels/office buildings

Technology Stack: - Localization: RTK-GNSS + visual/LiDAR fusion - Perception: Multi-sensor fusion (cameras, LiDAR, ultrasonic) - Planning: Traffic rule compliance + pedestrian avoidance - HRI: Pickup notifications, voice interaction, screen display

Indoor Delivery

Delivery robots in hotels/hospitals/office buildings must address special challenges: - Elevator interaction: Via IoT interfaces or robot cloud platforms to control elevators - Access control: Integration with building management systems - Multi-floor navigation: Cross-floor path planning

Warehouse Logistics Robots

Autonomous Mobile Robots (AMR)

Warehouse AMRs are among the most successfully commercialized service robot segments.

Platform Company Approach Deployment Scale
Proteus Amazon Robotics Fully autonomous, no retrofit needed Amazon warehouses
Kiva (original) Amazon Robotics Shelf transport (goods-to-person) Hundreds of thousands
Pick Geek+ P-series picking, M-series transport 500+ global projects
HAIPICK Hai Robotics Automated Case-handling Robot (ACR) Global warehousing
Locus Locus Robotics Collaborative picking North American warehousing
Quicktron Quicktron Tote/shelf transport Extensive domestic deployment

Warehouse Scheduling

Multi-AGV/AMR coordination scheduling is a core technical challenge:

Multi-Agent Path Finding (MAPF): Given \(N\) robots with respective start/end points, find collision-free paths minimizing total cost:

\[ \min \sum_{i=1}^{N} c_i(\pi_i) \quad \text{s.t.} \quad \pi_i \cap \pi_j = \emptyset, \quad \forall i \neq j \]

Common algorithms: - CBS (Conflict-Based Search): Plan independently first, then resolve conflicts - ECBS: Bounded suboptimal CBS, faster - Priority planning: Plan sequentially by priority; later-planned robots yield to earlier ones

Goods-to-Person vs Person-to-Goods

graph LR
    subgraph GTP["Goods-to-Person"]
        A1[Order dispatched] --> A2[AMR goes to shelf]
        A2 --> A3[AMR transports shelf/tote<br/>to workstation]
        A3 --> A4[Manual picking]
        A4 --> A5[AMR returns shelf]
    end

    subgraph PTG["Person-to-Goods + Collaboration"]
        B1[Order dispatched] --> B2[AMR goes to designated shelf]
        B2 --> B3[LED indicates pick location]
        B3 --> B4[Worker picks into AMR]
        B4 --> B5[AMR goes to next shelf or outbound]
    end

Food Service and Hospitality Robots

Food Delivery Robots

Food delivery robots are one of China's most successful service robot deployments:

Platform Company Features Deployment
BellaBot Pudu Robotics Cat design, multi-tier trays, interactive expressions 60+ countries globally
KettyBot Pudu Robotics Advertising screen + delivery Restaurants/malls
Servi BearRobotics US/Korean design North America/Asia restaurants
OrionStar OrionStar Voice interaction + delivery Hotels/restaurants
Peanut Series Keenon Robotics Multiple models, all-scenario Extensive domestic deployment

Core Technical Requirements: - Navigation in crowded dynamic environments (people, chairs, waitstaff) - Multi-robot coordination (avoiding corridor congestion) - Human-robot interaction (voice announcements, touchscreen ordering) - Stability (soup on trays must not spill)

Cleaning Robots

Consumer: Robot Vacuums

Robot vacuums are a mature consumer electronics category. Technology evolution:

Random collision -> Inertial navigation -> Laser SLAM -> AI vision

Major brands: iRobot Roomba, Roborock, Dreame, Ecovacs, Narwal

Coverage Path Planning

Coverage path planning aims to traverse all reachable areas while minimizing overlap and total path length.

Boustrophedon decomposition: 1. Decompose free space into convex sub-regions (cells) using vertical sweep lines 2. Execute boustrophedon (back-and-forth parallel line) scanning within each cell 3. Determine cell visit order (variant of traveling salesman problem)

Coverage target:

\[ \text{Coverage} = \frac{|A_{cleaned}|}{|A_{free}|} \times 100\% \]

Commercial robot vacuums typically require Coverage > 95%.

Commercial Cleaning Robots

Commercial settings (malls, airports, office buildings) use larger cleaning robots:

Platform Company Function Scenario
Whiz SoftBank Robotics Vacuuming Office buildings
Neo 2 Avidbots Floor scrubbing Malls/airports
CC1/CC3 Gaussian Robotics Scrub + sweep Large venues
Phantas Gaussian Robotics Outdoor sweeping Campuses/streets

Companion and Social Robots

Social Robot Design Principles

  • Non-verbal interaction: Eye contact, body language, facial expressions
  • Social distance: Following human social distance norms (intimate/personal/social/public)
  • Affective computing: Recognize user emotions and adjust responses
  • Appropriate anthropomorphism: Uncanny Valley effect — too human-like becomes unsettling

Representative Platforms

Platform Company Features Status
Pepper SoftBank Robotics Humanoid, emotional interaction, chest tablet Limited production
Jibo Jibo Inc. Desktop social robot, rich expressions Discontinued
Vector Anki -> DDL Micro desktop, AI voice Community maintained
Moxie Embodied Inc. Child companion, social-emotional Family-oriented
Loona KEYi Tech Pet-type companion, expression interaction Consumer-grade

LLMs Empowering Social Robots

Large language models have brought a qualitative leap to social robots: - Natural language dialogue: From preset scripts to open-domain conversation - Context understanding: Understanding conversation context and user intent - Personalization: Adjusting interaction style based on user habits - Multimodal: Comprehensive interaction combining speech, vision, and motion

Nav2 is the most commonly used navigation framework for service robots:

graph TB
    subgraph Nav2_Architecture["Nav2 Architecture"]
        BT[Behavior Tree<br/>Navigator] --> PLAN[Planner Server<br/>Global Path Planning]
        BT --> CTRL[Controller Server<br/>Local Controller]
        BT --> REC[Recovery Server<br/>Recovery Behaviors]
        BT --> SM[Smoother Server<br/>Path Smoothing]

        PLAN --> NAV[NavFn / Theta*<br/>Global Planning Algorithms]
        CTRL --> DWB[DWB / MPPI / RPP<br/>Local Planning Algorithms]

        MAP[Map Server<br/>Static Map] --> PLAN
        COST[Costmap 2D<br/>Cost Map] --> PLAN
        COST --> CTRL

        SLAM_[SLAM Toolbox] --> MAP

        LIDAR[LiDAR] --> COST
        DEPTH[Depth Camera] --> COST
        ODOM[Odometry] --> CTRL
    end

Costmap

Nav2 uses layered costmaps to represent navigation space:

  • Static Layer: Static obstacles from SLAM map
  • Obstacle Layer: Dynamic obstacles from real-time sensors
  • Inflation Layer: Inflates obstacles to ensure safety distance
  • Keepout Filter: No-go zone settings

Cost value \(c(x, y)\) meaning:

\[ c(x,y) = \begin{cases} 254 & \text{Lethal obstacle} \\ 253 & \text{Inscribed (within inflation)} \\ 1-252 & \text{Traversable but with cost} \\ 0 & \text{Free space} \end{cases} \]

Crowd Navigation

Special navigation needs for service robots — moving safely and naturally through crowds:

  • Social Force Model: Virtual forces modeling interactions between pedestrians and between humans and robots
  • DRL Navigation: End-to-end learned crowd navigation policies
  • Social Costmap: Adding social distance layers to costmaps
\[ \mathbf{F}_i = \mathbf{F}_i^{goal} + \sum_{j \neq i} \mathbf{F}_{ij}^{social} + \sum_w \mathbf{F}_{iw}^{wall} \]

where \(\mathbf{F}_{ij}^{social} = A_i \exp\left(\frac{r_{ij} - d_{ij}}{B_i}\right) \hat{\mathbf{n}}_{ij}\) is the social repulsive force.

AI Upgrades for Service Robots

  1. Foundation model integration: LLM/VLM enabling natural interaction and task understanding
  2. Multimodal perception: Fusing vision, speech, and tactile for comprehensive perception
  3. Cloud-edge-device collaboration: Cloud brain + edge inference + onboard control
  4. RaaS model: Robot-as-a-Service, on-demand rental rather than one-time purchase

References

  • Macenski et al., "The Marathon 2: A Navigation System", IROS, 2020
  • Choset, "Coverage of Known Spaces: The Boustrophedon Cellular Decomposition", Autonomous Robots, 2000
  • Helbing & Molnar, "Social Force Model for Pedestrian Dynamics", Physical Review E, 1995
  • IFR, World Robotics Service Robots Report, 2024

Related Notes:

评论 #