Mobileman, Construction Agile Goods Delivery Robot*

CHEN IMing, FENG Yue, LI Bingbing, TJU Hendra Suratno, SONG Hanyu

(Robotics Research Center, Nanyang Technological University, Singapore)

Abstract: The traditional automated guided vehicle (AGV) on goods delivery faces the challenges when task space expands beyond 2D plans. 3D environments such as uneven terrain, ramps, and staircase are typical in construction site. Thus, the key to introducing this technology into construction industry is to improve AGV’s stability and autonomous navigation ability in more complex three-dimensional environments. In this paper, mobileman, a novel tracked autonomous guide vehicle, is introduced. Compared with other construction robots, mobileman maximizes its load capacity on the basis of assuring accessibility. Furthermore, its modular designs and self-balancing platform enable it to cope with more complex challenging scenarios, such as staircase with 35-degree sloped staircase, while another modular design featured automated loading and unloading functionality. The mobile base specifications were presented in section two, and modular designs and exploration of the navigation system on construction site were illustrated in the rest of sections.

Keywords: construction robotics；intelligent robotic system；self-balancing platform；automated guided vehicle

0 Introduction

Automated guided vehicles (AGVs) have been widely used for goods delivery in areas, like exploration, manufacturing and logistics. Traditionally, these AGVs only need to deal with 2D navigation challenges in structural environments. However, in the construction site, 3D environments such as stairs and uneven terrain present unavoidable challenges. In recent years, as the increase of robot perception and intelligence, designing an unique AGV that can navigate and perform tasks in more challenging 3D environments has become feasible.

Because of its high slope and limited space, the staircase is the most challenging of the complex 3D environments in construction site. For years, stable stair climbing on the staircase has been a popular topic in robotics. leg-based[1], wheel-leg-based[2-3], and track-based mechanisms are the three basic types of mechanical structures used for stairs climbing. Among three types, the track-based mobile base is the most appropriate design for goods transportation for its stability and durability. Mobileman is a track-based mobile robot that is designed for stair climbing on standard prefabricated prefinished volumetric construction (PPVC) staircase and integrated with interchangeable modules and navigation system.

Unlike manual track-based stairs carriers, fully automated tracked-based stairs-climbing AGVs require an automatic balancing platform to keep the entire centre of gravity (CG) within the supporting area. Thus, a detachable self-balancing platform was developed. A loading and unloading module was designed and developed in addition to the self-balancing platform module. Both of the modules can be used to replace the default setting, which is a basket carrier on the top of the robot as shown in Fig.1. The loading and unloading module is a remotely controlled forklift with a maximum load capacity of over 120kg. In the future, we intend to create more interchangeable modules to increase the functions.

Fig.1 Mobileman running on 5 degrees ramp with 190kg payload

On site navigation of track-based robot can be the most challenging issue. Unlike wheel-based robots, the spin turn center of the robot can be vary when the CG of the entire system changes due to the varied payloads. In addition, the variation in ground surface conditions beneath the two tracks can make it slippery, causing traditional navigation to perform poorly in terms of accuracy. In such unstructured environments, however, changing road conditions are typical. In construction site, safety issue is always a top priority. The selection of automation level must be cautious. As a pioneer of exploring the AGVs in construction industry, we are attempting to integrate semi-automation to the robot, which is a pattern of mixed manual and autonomous operation, in order to quick connect with the industry.

This paper is organized as follows. Section Ⅱ illustrates the systematic framework design for the mobile base, including electronic components integration and software architecture, as well as the mobile base structural design that adheres to staircase requirements. Section Ⅲ introduces modular designs, such as the self-balancing platform and loading and unloading module. Section Ⅳ provides the information of the navigation system. The test results on site are given in the section Ⅴ, followed by a summary and future work in section Ⅵ.

1 Stair-climbing mobile base

In this section, we develop the systematic framework design for the mobile base, including the electronic components integration and software architecture, as well as the mobile base structural design that adhere to staircase requirements.

1.1 Systematic framework

1) Interchangeable modular system

The robot system is divided the mobile base system and the task-specific module at the top of the robot. The task-specific module is determined by the usage application, for example, carrying items up a staircase may necessitate the use of self-balancing module, automated goods transport may necessitate the use of a loading and unloading module, and object manipulation may necessitate the use of a robotic arm module.

These modules are easily interchangeable. Each module receives DC power from the mobile base system and, in most cases, has its own built-in controller. This controller exchanges information with the main computing system of mobile base via IO module, USB and/or Ethernet connection as explained later. The overview of this modular framework is shown in Fig.2.

Fig.2 Mobile robot and interchangeable modular system framework

2) Power management system

Various construction-related equipment and modules have different voltage and current requirements. Furthermore, these modules often consume high current and generate big electrical noises in the system, especially when the equipment is turned on and off. Loading and unloading module (24V/30A), grinder system (19V/20A) or robotic arm module (48V/25A) are just a few examples.

To avoid unexpected voltage decreases or surges, we divided the power management into two blocks. The first block is an isolated power management module, which controls different voltage levels at low current including 24, 12, 5V. This is an isolated and robust block that powers all sensors, computing and networking equipment, such as CPU, MCU, IMU, network switch, wireless access point, laser scanner, cameras.

The second block is the high current power block, or the high current interchangeable module. It is exclusively used to directly power the motor. To meet different requirements, our robot features redundant and heavy-duty cables and connectors. The robot has several empty DC/DC converter slots, which allows us to switch or add a generic DC/DC conveter module for different usage applications.

3) Mobile base system

The mobile base system comprises of one minicomputer system (CPU) that serves as the robot’s central brain, one microcontroller (MCU) that controls low latency devices, one dual-axis motion controller for close-loop control of the right and left track systems, and various sensors and peripherals.

The entire mobile base system and components is shown in Fig.3, which includes various sensors such as a 2D laser scanner for obstacle avoidance and localization, a 9-axis inertial measurement unit (IMU) for pose estimation, a distance sensing module connected to 6 ultrasound sensors for obstacle avoidance, a Bluetooth module for wireless joystick, and a wireless access point (WAP) for wireless robot control, etc.

Fig.3 Mobile base system and components

The motion controller and MCU unit are connected to the computer directly through RS232 and USB respectively, while the additional sensors are connected via a Network Switch or USB hub. Based on the computer’s input command, the motion controller executes a closed-loop position control and feeds back the current encoder readings. Along with the measurements from other sensors, the computer will use this reading to compute odometry and perform pose estimation.

USB, Ethernet, or IO Modules can be used to communicate between mobile base and interchangeable module. These three options will cover the majority of applications as well as the creation of future modular systems. An Ethernet network can connect a complex application such as a robotic arm module. CAN, Modbus, RS485, RS422 and RS232 device modules can be connected to the USB or Ethernet via an adaptor. Finally, simple motor control application and sensor reading can be controlled directly using analog and digital input-output via the MCU. This MCU is specialized to the modular system and operates at a low latency.

1.2 Staircase standards & structure design

1.2.1Staircase standards

Due to the limited land and the need to speed up development, Singapore construction companies are now using PPVC, which stands for prefabricated prefinished volumetric construction. So, Singapore Building and Construction Authority (BCA) establishes a PPVC staircase standard with a width of 1m, a riser of 165,175mm, a tread of 250mm and a nosing of 25mm, a toe-to-toe distance about 305mm, and an angle of approximately 33, 35 degrees. The intermedium platform is about 1 200mm width and 2 150mm long. The PPVC staircase standard is shown in Fig.4.

Fig.4 Standard PPVC staircase from BCA database

1.2.2Structure design

In most construction sites, two types of mobile base are used wheel type and track vehicle. The passive track vehicles are better suited to the dynamic environment than wheel type AGVs. The passive crawler track, especially while climbing a staircase, will provide a continuous contacting surface for the mobile base, as it moves up and down the staircase, reducing the wavy and damping. In addition, climbing performance of various types of rubber track varies. In our experiments, the bar type of rubber belt will provided the optimum performance for staircase climbing as illustrated in Fig.5.

Fig.5 Bar type rubber track

To operate in a stairwell, the size must be designed to fit the stairwell, and the belts must contact three stair toes to provide ascending stability. Thus, when AGV is moving, the minimum number of contacting toes must be two. Our mobile base will not be able to turn around, as if we need it to carry as many cargos as possible. To make 90-degree turn on platform, the length of mobile base cannot exceed 910mm, and the width cannot exceed 650mm. To give a stable climbing performance, the flat bottom surface of the rubber track must be 610mm at minimum (to make sure the base could land on three toes at the same time). When the torque of the base is sufficient to climb the staircase, the height of the top of the belt to the bottom must be at least 410mm. To reduce the wavy and damping when climbing the staircase, the motors will be situated in front of the base, and no spring suspension will be applied to our base.

2 Modular design

This section introduces two modular designs of the MobileMan. One is the self-balancing platform, and the other is loading and unloading module. Both modules have been developed and tested with payloads. The concept of self-balancing platform was validated with the initial smaller-scaled prototype, while the loading/unloading module was tested on site with the full size second prototype.

2.1 Self-balancing platform

There are a variety of powered track-based stairs carriers on the market, all of which require human assistance and control. For example, in 1990, US filled a patent for a stair-climbing wheelchair carrier. XSTO, and Pianoplan both sell commercial products, such as stair climbers for goods delivery. All of these manual devices share a common mechanism: a one-degree-of-freedom balancing platform between the tracked base and the cargoes. The primary purpose of this mechanism is preventing the climber from flipping over when on the stairs. This type of machine requires human assistance at all time for safety and stability.

Unlike manual track-based stairs carriers mentioned above, fully automated tracked-based stairs-climbing AGVs requires automatic balancing platform to keep the entire centre of gravity (CG) within the supporting area. In paper[4], we introduced this 2-Dof self-balancing platform, which uses an IMU sensor and PID control to keep the platform’s surface horizontal to the ground. The topological kinematics of the self-balancing platform are depicted in Fig.7. It has workspace in both the roll and pitch directions, with output rotational ranges of [-15°,15°] and [0°, 40°] correspondingly. This parallel mechanism is driven by two SSPS-105 motors, each with a maximum output torque 37 Nm. According to the results of force analysis, the maximum payload can theoretically be up to 35.8kg. As illustrated in Fig.6, the first prototype mobile base was equipped with this module and was capable of carrying 25kg cargoes up a 25-degree sloped stairwell. This parallel mechanism depicted in the picture is a small-scaled variant in order to be compatible with this smaller-scaled chassis. In the near future, the full-size mechanism will be used to a fully autonomous stairs climbing tracked-based AGV with a higher payload of up to 100kg, mainly used for the construction materials transportation.

Fig.6 Mobileman first small-scaled prototype, equipped with self-balancing platform, carrying 25kg payload

Fig.7 Topological kinematics of the self-balancing platform

2.2 Loading/unloading module

The loading and unloading module was inspired and designed based on the structure of a forklift typically used on construction sites. It has two degrees of freedom and two joints, one rotating and one prismatic. The workspace of rotating joint is [90°, 180°], while the workspace of the prismatic joint [0mm, 500mm]. Unlike traditional forklifts, it has a substantially greater workspace in the rotational joint, allowing the mechanism to be completely folded to maximize the robot’s accessibility, while minimizing the robot’s footprint. The mechanism is substantially smaller than a regular forklift. From the top view, it measures 805mm×610mm, when fully folded. The forks are 500mm long. The cargoes that are usually targeted are tiny products found on construction sites, such as tiles boxes, paints buckets and rubbish bins. The payload for the aluminum prototype is 120kg, however, if the material changes to steel, the payload can be increased to 200kg.

The module is designed as a remote control, and operators can control the two joints independently by a joystick. The loading operations are depicted as pictures a to c in Fig.4, and the unloading processes are the reversed actions. The robot in picture d was carrying 60kg tiles boxes while travelling on a 6° slop. The following are the loading processes:①Using the mobile base, aim the gap underneath the cargoes. then, control the mechanism until it is fully extended;②Slowly approach the cargoes with the robot, ensuring the cargoes are close enough to the vertical bars; ③Raise the cargoes and set the rotational joint to the correct angle.

3 Navigation

To increase the autonomy of the robot, a localization and navigation system is added to assist the control of the robot in both ground and staircases. The robot has two modes: a fully autonomous mode for navigating on flat grounds and a wall-following mode for moving around curbs and staircases. The system is built in such a way that current state-to-the-art 2D navigation methods can be used, and 2D sensors can minimize the overall sensor costs.

3.1 Localization and mapping

Localization of a robot is the prerequisite for navigation to gather pose information from multiple sensors. To create and update maps in an unknown environment while simultaneously track the position and posture of the robot, well studied SLAM algorithms are available as opensource software.

Fig.8 Mobileman second prototype, equipped with loading/unloading module, picked up and carried 60kg payload

The localization hardware consists of one inertial-momentum unit (IMU), one 2D laser scanner (20m working range), six ultrasonic sensors, and wheel motor encoders. The IMU is used to measure the tilt angles of the robot and serves as the primary information source for determining whether the robot is on a staircase. It also serves as a data source to correct the odometry of the robot. Wheel encoders are used as the main source to calculate the odometry of the robot, and when combined with the IMU and the Kalman Filter, a more consistent and accurate odometry can be obtained. Obstacle avoidance uses ultrasonic sensors as safety sensor. The standard adaptive Monte Carlo localization (AMCL) approach is used for robot localization, and the laser scanner mentioned above is the main source of sensor input.

Mapping, Hector SLAM and Cartographer are the three popular SLAM algorithms that are well developed and deployed in commercial robots. GMapping uses Rao-Blackwellized Particle Filter, and the number of particles for the filter and number of iterations in the scan matcher can greatly impact on the mapping result. However, the processing capability of the robot’s main controller (Intel Cherry Trail Z8350 Processor @ 1.44Ghz) is relatively low, limiting the use of GMapping.

Comparing with Hector SLAM, cartographer utilizes odometer data to generate a superior global map using closure loop submap matching algorithms. When comparing the size of a construction site with the working range of a laser scanner, this approach for mapping generation is advantageous in a construction environment.

3.2 Navigation with size constrains

The topic of navigation in a 2D planetary environment has been extensively studied. In our application, in major part of the construction site, open space is guaranteed for the robot to pass. A Navfn global planner is used to generate point-to-point global path, and a DWA (dynamic windows approach) local planner is used to generate a real-time velocity profile and perform obstacle avoidance. However, DWA can work best in a differential drive robot with zero turning radius, which will not be possible in our tracked vehicle because the turning center is not fixed. Furthermore, in a constrained environment such as in the stairwell, DWA cannot work well in the limited configuration space. Thus, a mechanism to replace the velocity command sent by local planner is developed.

The robot is aware of its status (i.e., whether it is climbing staircases or on a plane) assisted by IMU-measured robot tilt angle (an IMU capable of sensing gravity acceleration is required), and the extra information provided by construction standard (the staircase slope). The movement can be projected to a narrow, lengthy corridor in either scenario.

When the insufficient space restricts the robot to make a full 180-degree turn in the long tight hallway, the heading angleθbecomes critical. Any turning will drastically impair the maneuverability of the robot. The odometry is inaccurate due to the slippery of tracks on staircase, making determining the heading angle difficult. The unevenness and hollowness of the walls around staircases, as shown in the far left in Fig.9, makes ultrasonic sensors readings unreliable. To get the accurate heading angle, each laser scan is divided into either parts, each having a 45degrees angle, and the heading angle is then obtained by

Fig.9 The heading angle constrain by the geometry of staircases

θ=arctanβ

whereβis given by

Each pair of (x,y) represents one point in the laser scan,αis the distance from the laser scanner to the surface. The value with the minimumQwill be chosen since it contains the least amount of distortion due to various sources.

Because of the length of the robot and the width of the staircase, the heading angle at a point is limited to avoid collision, and the maximum heading angle constrained by one side of the wall is then found by

θmax=arctan2(d-0.5w,0.5l)

Wherewis the width of the robot,lis the length anddis the distance from the laser scanner to the wall. Perpendicular walls in the turning point and parallel walls in the straight staircases are found in the staircases. Each pixel indicates a point inxandyaxis. The color indicated the maximum angle of the robot, with red being the high and blue being the low.

The optimum path in the turning point of staircases can therefore be generated by performing a neighborhood KD tree search form the entry point of the turning platform to the exit point. The heading angle heatmap can be generated offline depending on the design of the building and the size of the robot. The center is the optimum location for exist turning.

4 Site test results

The robot was tested for endurance and functionality in two different construction sites for three days. One of the two sites has a staircase width of 1.4m, while the other one has a staircase width of 1.1m. Both have slopes with a 5-degree slope ratio for large vehicles..

The robot with empty load and 50kg load successfully climbed the staircases in both sites (with slope of 35 degree) in two different layouts. The average climbing speed is 0.1m/s. After six rounds of climbing, mild wear-off of the rubber track was observed. Even with the assistance of the auto-balancing system, significant vibration still observed when the robot lands on the ground. It was also found that maneuvering the robot was more difficult on a narrower staircase. The problem was exacerbated by any payload. Besides, when robot was driven from staircase to the horizontal ground, the last step landing impact is noticed .

After that, payload tests were performed. In a plan ground and 5-degree slope, the robot runs from an empty cargo to a maximum payload of 200kg. A summary of the average speed is listed in Table 1. It is concluded that even with two 800W motors, the robot could not reach the targeted velocity when the burden exceeded over 140kg, however, the robot can still attain the same speed even with a payload of 200kg. This indicates that for the control system to work properly, the payload should be between 70, 140kg. After 4 hours of payload testing, the motor driver and gearbox overheat, implying that a lower payload should be used in long-term operations.

Table 1 Nominal speed of the robot kg

5 Conclusion and future works

5.1 Conclusion

In this work, we constructed a tracked-base robot that is equipped with numerous modules to perform multiple-functional products delivery tasks and can be deployed in diverse construction environments due to the specified accessibility and stability of the structure. Our robot is equipped with an affordable navigation system that operates on both the ground and the staircase, allowing it to operate independently. On site tests were conducted to demonstrate that the proposed products delivery robot could be successfully applied to practical stair climbing scenarios with a specific payload. Durability and maximum payload tests were also conducted. In the most difficult stair climbing scenario, however, instability was still observed due to the size limits.

5.2 Future works

1) Because the current prototype size is constrained by the PPVC staircase standard, which is extremely narrow and steep. For the next generation, we intend to design a larger version of Mobileman that can still working in most construction environments with greater payload capacity and stability.

2) For the last step of stair climbing, a landing system will be developed to cushion the impact of an unexpected drop off.

3) Distance control will be purposed, allowing operators to monitor the status of the autonomous robot and switch to manual control mode as needed. Cameras and ultrasonic sensors will be installed on each robot. A 5G network will be used to transmit data collected from sensors to the front end. Workers, on the other hand, can operate the robots by observing the visualized data such as 360 degrees live video streaming.