Source-linked AI summary
The Design of Stretch: A Compact, Lightweight Mobile Manipulator for Indoor Human Environments
Charles C. Kemp, Aaron Edsinger, Henry M. Clever, Blaine Matulevich
TL;DR
Indoor mobile manipulators can perform varied tasks, but their adoption has been limited, and reducing their size, weight, and cost risks constraining capabilities. Stretch combines a differential-drive base, lift, telescoping arm, and wrist and gripper extensions; anthropometry, stability models, and real-home tasks support the design. The authors report a growing user community around the commercial Stretch RE1 and suggest it indicates increasing adoption.
Problem
Mobile manipulators have potential for varied indoor tasks, but have been used primarily in research labs, with widespread home and office use not yet realized.
Source: §I · p. 1Method
Stretch combines a differential-drive mobile base, lift, and telescoping arm for Cartesian end-of-arm motion, with wrist and gripper extensions, and justifies the design using anthropometry and static-stability models.
Source: §I · p. 1, C. The Four Main Joints · p. 2, E. The Wrist and the Stretch Dex Wrist… · p. 3Results
Takeaways & Limitations
The authors suggest that a growing community using the commercially available Stretch RE1 indicates their design is increasing mobile-manipulator adoption.
Source: §V · p. 6
Abstract
from arXiv · showhide
Mobile manipulators for indoor human environments can serve as versatile devices that perform a variety of tasks, yet adoption of this technology has been limited. Reducing size, weight, and cost could facilitate adoption, but risks restricting capabilities. We present a novel design that reduces size, weight, and cost, while supporting a variety of tasks. The core design consists of a two-wheeled differential-drive mobile base, a lift, and a telescoping arm configured to achieve Cartesian motion at the end of the arm. Design extensions include a 1 degree-of-freedom (DOF) wrist to stow a tool, a 2-DOF dexterous wrist to pitch and roll a tool, and a compliant gripper. We justify our design with anthropometry and mathematical models of static stability. We also provide empirical support from teleoperating and autonomously controlling a commercial robot based on our design (the Stretch RE1 from Hello Robot Inc.) to perform tasks in real homes.
I. INTRODUCTION
Mobile manipulators could serve many useful indoor tasks, but widespread home and office use has not yet been realized. Stretch addresses this gap with a compact, lightweight, lower-cost design intended to preserve useful capabilities, developed through mechanical modeling and iterative prototyping.
I. INTRODUCTION
- Reducing size, weight, and cost is proposed as a way to improve adoption, but smaller scale can restrict stable workspace, loads, and task feasibility.Larger robots have greater swept volume and mass, while higher cost can make applications infeasible and deter production.Source: §I · p. 1
- The design process combined mechanical models with iterative prototypes tested on varied tasks, including eight prototypes tested in an Atlanta home from 2017 to 2020.Two earlier prototypes were tested at Georgia Tech from 2016 to 2017.Source: §I · p. 1
- Stretch is reported significantly smaller, lighter, and less expensive than prior mobile manipulators with comparable capabilities.The paper presents the design, its justification, and empirical evidence for its efficacy.Source: §I · p. 1, Table I · p. 1
II. THE DESIGN OF STRETCH
- The design takes minimalist inspiration from the Roomba, whose compact, lightweight, low-cost form was matched to autonomous floor cleaning in homes.The paper links those characteristics to the Roomba's body, sensors, and computation being matched to its task and environment.Source: §II · p. 1
A. The Structure of Indoor Human Environments
Stretch is designed around the Cartesian structure of indoor environments and the human dimensions associated with reaching, navigation, and perception. Anthropometric comparisons position the robot as a deconstructed, task-matched alternative to a humanoid form.
A. The Structure of Indoor Human Environments
- Indoor spaces center on horizontal planes and vertical surfaces; accessible locations and objects tend to be reachable by people and visible from head height.People, children, older adults, people with disabilities, and pets also create safety challenges, including falls, pinch points, and unpredictable motion.Source: A. The Structure of Indoor Human… · pp. 1–2
- Stretch uses static stability and predictable Cartesian motions; its 0–110 cm floor-to-reach range covers standard countertops up to 92 cm.Its 23 kg mass is above single-person lifting guidelines, but the robot can be lifted by two people or rolled by one.Source: A. The Structure of Indoor Human… · p. 2
B. Matching Human Dimensions
- Anthropometry frames Stretch as “robotic cubism,” matching key human dimensions without adopting a humanoid root.The figure compares reach to the ground and above countertops with hip width and arm length.Source: B. Matching Human Dimensions · p. 2, Figure 2 · p. 2
- 34 cm stowed width is below the 50th-percentile hip widths reported for adult females (37.1 cm) and males (36.1 cm).The text notes that paths wider than human hips facilitate comfortable walking and are likely common in human environments.Source: B. Matching Human Dimensions · p. 2
- 71 cm reach from the base is similar to 50th-percentile arm lengths for adult females (67.3 cm) and males (72.6 cm); the 131 cm camera height falls between reported seated and standing eye heights.The eye-height comparisons are given separately for adult females and males.Source: B. Matching Human Dimensions · p. 2
- 112 cm fingertip height is close to 95th-percentile right shoulder heights of female (109.7 cm) and male (113.4 cm) wheelchair users.The paper connects this reach to facilitating assistance for people with disabilities.Source: B. Matching Human Dimensions · p. 2
C. The Four Main Joints
Stretch uses four main actuators to combine mobility with Cartesian end-of-arm motion while limiting actuator count. The design relies on the mobile base for a degree of manipulation that the lift and telescoping arm alone cannot provide.
C. The Four Main Joints
- Four actuators produce Cartesian end-of-arm motion and flat-floor mobility, reducing actuator requirements that affect robot weight, complexity, and cost.The actuators comprise the telescoping arm, lift, and two differential-drive wheels.Source: C. The Four Main Joints · p. 2
- A fixed-length arm with three omniwheels would either increase width with a long arm or lose human-length reach with a short one.Additional omnidirectional-drive actuators could also complicate traversal of thresholds, rugs, cables, and tile.Source: C. The Four Main Joints · p. 2
- The base moves the end of the arm laterally, the lift moves it vertically, and the telescoping arm moves it in and out, yielding an identity Jacobian across the full workspace.The corresponding manipulability ellipsoid is an axis-aligned sphere.Source: C. The Four Main Joints · pp. 2–3
- The mobile base is critical because the lift and telescoping arm position the end of the arm only within a vertical plane.The paper contrasts this with manipulation using a stationary mobile base and with mounting a conventional serial manipulator on a mobile base.Source: C. The Four Main Joints · p. 3
- A lightweight telescoping arm, including materials such as carbon fiber, supports long reach with low weight and limits its influence on the robot's center of mass.On flat floors, only the lift actuator works directly against gravity.Source: C. The Four Main Joints · p. 3
D. Two Modes of Operation
Stretch is described through navigation and manipulation modes, which differ in the robot's orientation and use of its mobile base. Moving between them requires selecting a task-relevant position where the target remains within reach.
1) Navigation Mode:
- Navigation mode retracts the arm within the base footprint, letting the robot travel as a conventional differential-drive mobile robot.The lift can lower the arm and the robot's center of mass, increasing stability.Source: 1) Navigation Mode · p. 3
2) Manipulation Mode:
- Manipulation mode treats the arm-extension direction as the front and uses the base for Cartesian end-of-arm motion, including rotation around a vertical axis.The base can also perform curvilinear motions in this mode.Source: 2) Manipulation Mode · p. 3
3) Mode Switching:
- Switching modes usually follows navigation to a task-relevant location, and requires choosing a position where the target will be reachable after the switch.The transition often involves rotating in place, lifting and extending the arm, and turning the head toward the tool.Source: 3) Mode Switching · p. 3
E. The Wrist and the Stretch Dex Wrist Accessory
Stretch’s wrist stows tools within the base footprint and adds dexterity; the optional Dex Wrist adds pitch and roll. The compliant gripper adapts an affordable assistive grabber for robot use.
E. The Wrist and the Stretch Dex Wrist Accessory
- The 330° yaw joint stows tools within the base footprint for navigation and adds dexterity for camera viewing and grasping elongated objects.The wrist adds a fifth DOF, providing redundancy for rotations of the mobile base.Source: E. The Wrist and the Stretch Dex Wrist… · p. 3
- The Dex Wrist adds pitch and roll to yaw, enabling tasks such as pouring and operating door knobs.Its fingertip center reaches approximately 130 cm; the four main joints provide gross motion, while the wrist and tools perform fine movements.Source: E. The Wrist and the Stretch Dex Wrist… · p. 3
F. The Stretch Compliant Gripper
- Six assistive-robotics lab members informally evaluated ten grabbers, leading to selection of the VIVE Reacher and Japanese Reacher Grabber.The candidates were tried with objects relevant to assistive robotics.Source: F. The Stretch Compliant Gripper · p. 3
- The Stretch Compliant Gripper adapts the VIVE grabber with actuator kinematic and torque feedback and significantly more grip force than the original human-operated grabber.A 90° mount and hook support door and drawer operation; the standard gripper tilts down for grasping from above or the side without a pitch joint.Source: F. The Stretch Compliant Gripper · pp. 3–4
G. Stretch’s Sensors
Stretch uses actuator current sensing for haptic load sensing and contact detection. Its cameras, range finder, microphones, and inertial sensors support perception, navigation, and safety functions.
G. Stretch’s Sensors
- Actuator current sensing detects applied loads, and the lift and arm’s prismatic joints support contact detection and autonomous drawer opening.The robot extends to contact a drawer surface, lowers to contact the handle, then retracts.Source: G. Stretch’s Sensors · p. 4
- The pan-tilt RGB-D camera can look along the travel direction, at the end-of-arm tool, or down at the base.Rotating the camera increases vertical field of view, capturing depth imagery ahead and near the base while navigating.Source: G. Stretch’s Sensors · p. 4
- A base-mounted laser range finder supports navigation, while the IMU can support navigation and tilt detection to avoid tipping and toppling.The robot also has a microphone array for speech recognition and sound localization, plus wrist accelerometers for bump sensing.Source: G. Stretch’s Sensors · p. 4
III. MODELING DESIGN TRADEOFFS
The models frame a tradeoff between reducing robot size and weight and increasing workspace and applied loads. Planar static-stability models relate payload and push/pull forces to reach, robot mass, base width, arm height, and center-of-mass position.
III. MODELING DESIGN TRADEOFFS
- The design trades smaller size and weight against greater workspace and applied loads, which the authors analyze with closed-form static-stability models.The models represent stability across the robot’s full workspace.Source: §III · p. 4
A. Planar Models of Static Stability
- The planar tipping models assume static equilibrium with all but one wheel losing ground contact.They represent the robot just before tipping under task-relevant loads.Source: A. Planar Models of Static Stability · p. 4
1) Maximum Payload:
- Maximum payload is proportional to robot mass and base width, but inversely proportional to arm reach.These relationships favor higher payload while also motivating less robot mass, a narrower base, and farther reach.Source: 1) Maximum Payload · p. 4
2) Maximum Pulling & Pushing Forces:
- Maximum pulling force increases with robot mass and base width but decreases with arm height; arm extension reduces maximum pulling force and increases maximum pushing force.The push-pull difference depends on the robot’s center-of-mass position relative to the drive wheels, which changes with telescoping-arm configuration.Source: 2) Maximum Pulling & Pushing Forces · pp. 4–5
B. Static Stability with a Triangular Support Polygon
Stretch’s triangular support polygon changes the moment arms used to model tipping, with a simplifying assumption that the center of mass is equidistant from the drive wheels. The resulting models describe payload and force limits across the workspace and are compared with measurements.
B. Static Stability with a Triangular Support Polygon
- The model accounts for tipping over the sides of Stretch’s three-wheel support triangle and assumes the center of mass is equidistant from the drive wheels.The center-of-mass moment arm is the perpendicular distance to the sides of the triangle.Source: B. Static Stability with a Triangular… · p. 5
- The triangular geometry changes payload and force moment arms; under the center-of-mass assumption, modeled maximum pushing and pulling forces have equal magnitude.The model also estimates backward pushing force and describes the Cartesian workspace by its height, depth, and width.Source: B. Static Stability with a Triangular… · p. 5
- Figures 7 and 8 compare model predictions with measured tipping loads, with Figure 8 distinguishing fully retracted and extended arm configurations.Figure 7 shows average worst-case measurements; Figure 8 shows measurements for both arm configurations.Source: Figure 8 · p. 5, Figure 7 · p. 5
C. Other Design Considerations
The design keeps elevated mass low while balancing toppling risk against reach: widening the base can increase full arm extension, while rolling the paused robot can increase toppling risk.
C. Other Design Considerations
- A lightweight mast, carbon-fiber telescoping arm, and small head reduce elevated mass, while the batteries and three main actuators sit in the base.The passage identifies a higher center of mass as increasing toppling risk from acceleration, bumps, thresholds, and ramps.Source: C. Other Design Considerations · pp. 5–6
- Widening the mobile base by i cm allows full arm extension to increase by up to n i cm, where n is the number of telescoping elements.This relation is stated while ignoring other factors.Source: C. Other Design Considerations · p. 6
- When paused, Stretch can be tilted over its drive wheels and rolled, which may improve ease of use but increase toppling risk.Figure 9 depicts teleoperated tasks 1–3 and 18–19, and autonomous tasks 10–12.Source: C. Other Design Considerations · p. 6, Figure 9 · p. 6
D. Measurements & Model Predictions
Measurements closely match the predicted worst-case payload at maximum reach, and tests confirm that static stability constrains the robot’s capabilities. The authors report that stable payloads and forces suffice for several assistive tasks.
D. Measurements & Model Predictions
- 3.46 kg measured worst-case payload at 0.693 m reach closely matches the model’s 3.47 kg prediction.The measured average was based on five trials with a standard deviation of 0.065 kg.Source: D. Measurements & Model Predictions · p. 6
- With force limits disabled, Stretch pulled with more than 70 N and lifted more than 60 N (> 6.12 kg), confirming static stability limits its capabilities.The authors did not attempt maximum achievable forces to avoid damage.Source: D. Measurements & Model Predictions · p. 6
- Stable forces and payloads suffice for reported assistive tasks, including opening many drawers and cabinet doors (< 20 N pulling), face wiping and shaving (< 10 N pushing), and lifting prioritized objects (< 1.2 kg).The tasks are described in cited literature; Table II is titled as a list of tasks performed by Stretch.Source: D. Measurements & Model Predictions · p. 6, Table II · p. 6
IV. TESTING THE DESIGN WITH A VARIETY OF TASKS
Stretch was tested on a variety of tasks, including teleoperated and autonomous tasks in real homes. The conclusion presents the design as a compact, lightweight, lower-cost manipulator and points to a growing user community as evidence of increasing adoption.
IV. TESTING THE DESIGN WITH A VARIETY OF TASKS
- Except for shelf-picking and inspection tasks, all listed Stretch tasks were performed in real homes; autonomous tasks used only onboard sensors and computation.The task videos and code are reported as available online.Source: §IV · p. 6
- In related work, 21 people, including three people with disabilities, remotely operated Stretch for kitchen tasks.Source: §IV · p. 6
V. CONCLUSION
- The conclusion says a growing community using the commercial Stretch RE1 suggests the design is increasing adoption of mobile manipulators.The paper supports its design with anthropometry, mechanical stability models, and tasks in real homes.Source: §V · p. 6