A Servicing Rover for Planetary Outpost Assembly

Medina, A. (1); Pradalier, C. (2); Paar, G. (3); Merlo, A. (4); Ferraris, S. (4); Mollinedo, L. (1); Colmenarejo, P. (1);Didot, F. (5)

(1) GMV, c/Isaac Newton, 1 28027 Tres Cantos – Spain, {amedina,lmollinedo,pcolmena}@gmv.es

(2) ETHZ-ASL, Tannenstrasse, 3 8092 Zurich - Switzerland, cedric.pradalier@mavt.ethz.ch (3) JR, Steyrergasse 17-19, 8010 Graz – Austria, gerhard.paar@joanneum.at

(4) TAS-I, Corso Marche 41, 10146 Torino – Italy, {andrea.merlo,simona.ferraris}@thalesaleniaspace.com (5) ESA/ESTEC, Keplerlaan 1, 2200 AG Noordwijk – The Netherlands, frederic.didot@esa.int

ABSTRACT

Robotic and human exploration of Moon and Mars, as recently envisaged by international Space Agencies, raise new objectives for mobile planetary devices in terms of mobility, autonomy and interaction with other space assets, including humans.

To prepare for such activities, the Eurobot Ground Prototype (EGP) Project was awarded by ESA to an industrial consortium lead by Thales Alenia Space Italia. Within this project, GMV is responsible for the development of the mobile platform “EGP-Rover” with Technical University of Zurich (ETHZ) and Joanneum Research (JR) from Austria as subcontractors.

This paper describes such EGP-Rover mobile platform as a candidate robot for future Lunar Outpost assembly activities.

1. INTRODUCTION

In support to human exploration, ESA has studied the Eurobot system [1][2][3] , a robot assistant to manned EVA sorties. Eurobot is a humanoid like robot; at first designed to service the International Space Station, it had a torso equipped with two arms, a single leg and a head camera. Regarding further steps of human exploration, there is the additional need to study and test the Eurobot derived technologies in support to human activities on-surface.

A fully functional Eurobot Ground Prototype in an “on-surface” configuration named EGP-Rover will be therefore developedand tested for early evaluations of functions, human-robot cooperation capabilities and relevant enabling technologies.

The goal of this platform is to achieve a highly-accurate mobile system able to manoeuvre in short distances to transport an astronaut or a payload to a target location.

The EGP-Rover is a mobile platform sized 1.5m x 1 m, aiming to assist astronauts in the construction and maintenance of a lunar or Martian base station. This platform (see Figure 1), considered as an on-Earth

technology demonstrator, weighs 880 Kg and has a payload capability of 400 Kg. By using space qualified components it has been demonstrated that the weight of the platform could be reduced by a factor of 1.5.

Figure 1 . EGP-Rover moving around the Lunar Lander mockup.

Such mobile platform must reach a target point at a distance of 10 m with a localization accuracy of 10cm and a heading accuracy of 2 degrees. Several solutions were considered to achieve the required high manoeuvrability and corner reachability (see Figure 2). In order to reduce mechanical complexity, a 4-wheel rover was designed with only two steerable wheels.

Figure 2. Reachability analysis for 60º corners.

The rover rotation point located at the middle point of the front wheel axis was aligned with the centre of the Eurobot arm support. For this reason, the final rover design resulted in the two rear wheels being independently steerable from 0 to 90 degrees. Therefore, the rover is able to drive on tight curves and to rotate on the spot.

Finally, the EGP-Rover was also designed to stay stable on rough outdoor terrain and to overcome obstacles of 20cm height. For such purpose, a passive rotation axis was included between the front and the rear part of the rover (see Figure 3).

Figure 3. EGP-Rover climbing obstacles.

Autonomous navigation is based on sensor data fusion through extended Kalman filtering from the following devices: low-cost inertial measurement unit (IMU), mechanical odometers, and a stereo-vision camera system. Slippage is detected through the comparison of individual wheel encoders and IMU data offers the baseline solution between rover way points.

The Rover Vision System (RVS) is mounted on top of the rover to provide a panoramic stereo view using a pan-tilt unit (PTU). For initial detection of the EGP position and pose, it uses artificial (coloured or specially shaped) targets mounted on infrastructure (e.g. the Lander). Initial high-resolution mapping results in a Digital Elevation Map (DEM) [3] and derived products such as slope map, hazard map and roughness map. These maps are analyzed using a cost function which delivers an optimum path (in terms of safety and cost) from the current EGP position to the desired target location.

This initial guidance solution is supplied by a path-planning A* algorithm where the rover is modeled as a single point and obstacles are enhanced within the navigational map. The trajectory is refined through cubic splines adjusted by the non-holonomic rover grid. During motion, the RVS provides near real-time hazard maps (again based on local DEMs) and updates the EGP position and pose every few meters by visual comparison of the initial DEM texture with the current image (“surface-relative orientation”). This process also refines an existing DEM, resulting in a larger, accumulated DEM version.

Steering rover control is done using an Ackerman model [6] where a guiding point is in the middle of the front axis and wheels steering are at the rear axis. In this model, an estimated linear speed acts as the control variable for linear traction over the 4 wheels, and a virtual angle commands the steering of the 2 rear wheels as shown in the next figure:

Figure 4. EGP-Rover Ackermann rear-wheels steering.

This Ackermann model allows having a rotation on-the-spot over the middle point of the front axis aligning the wheels at specific positions. Encoder accuracy and gear ratio over the DC brush-less motors is high enough to achieve wheel positioning accuracies in the range of mm. At the lowest level, the motor control is implemented using standard EPOS motor controller from Maxon AG, interfaced through a CAN bus linking all the motor controllers on the robot.

The structure of this paper is as follows: Section 2 covers the locomotion subsystem, Section 3 addresses the stereovision subsystem (RVS) while Section 4 describes the navigation system (RGNC). Finally, Section 5 presents final conclusions.

2. LOCOMOTION SUBSYSTEM

2.1. Chassis design

The main dimensions of the locomotion platform Figure 5 are listed hereafter: bounding box volume of 2246 x 1580 x 1505 mm, track width: 1330 mm (rear) and 1041 mm (front); wheel stance: 1511 mm and a ground clearance of 281 mm.

Figure 5. EGP-Rover top and front view.

Mechanically, the proposed design is optimized for motion on rough terrain (for instance, sand with rocks of various sizes but smaller than 20cm). The following details should be noted:

Articulated chassis: a 4-wheel vehicle without suspension is naturally hyper-static. As a result, if one wheel is driving on a rock, the three other wheels cannot stay on a flat surface and one of these three must be in the air. This creates an instable situation and a loss of traction. Two solutions can be envisaged to take care of this issue: either to have a suspension system or to have a freely rotating joint between the two front wheels and the two rear wheels. The solution based on a rotating joint was finally implemented due to the simpler resulting mechanics.

The articulated chassis assumes that the vehicle will really be used in rough terrain situation. If this is not the case, a simpler design could be made by removing the free joint. Small obstacles could still be passed using the natural flexibility of the wheel tires and the fact that the 4 wheels are actuated independently.

Step climbing: Chassis design allows climbing obstacles of 0.2 m height either with one single wheel (asymmetric obstacles) or with both wheels left and right simultaneously (see Figure 6).

Figure 6. Rover symmetric obstacle climbing.

Rotation on the spot: the ability to rotate around the main chassis axis requires all wheels to be steerable. As the two robotic arms must be located in the front part, the rover design took into account this requirement by providing the ability to rotate around the centre of the front axis, which could also be the origin of the arms frame of reference depending on the final desired arm configuration.

This design can be achieved by having only the rear wheels steerable at +/- 66°, as illustrated in Figure 7. On the other hand, this requires a steering angle of approximately +/- 120 degrees, which is only feasible if the steering motors are located above the steering wheel. Another advantage of this design is that a rotation around the arms attachment is likely to provide a more practical motion when doing assistive tasks with the astronaut.

Figure 7. Rotation on spot motion (left) and rear wheel steering mechanism (right).

2.2. Stability Analysis

Stability is ensured as long as the gravity vector is included in the hull of the supporting points. The gravity vector starts from the COM (Centre of Masses) and points down vertically. The supporting points are located where the weight of the system is supported. The hull is the shape drawn by these points. This vector is replaced with a cone. The whole cone must be included in the hull. The computation of the maximum rover slope stability is done using a purely geometrical relationship because of the simplicity of the rover structure.

The rover is split in front and rear part, and stability is checked for both parts independently depending on the respective centres of mass. Then, the overall stability is confirmed with the rover’s footprint and centre of mass.

The rover is considered to remain stable as long as the projection of the centre of mass along the gravity vector on the ground plane remains within the polygon defined by the support points (wheels or middle joint). For 360° heading stability analysis, the projection vector defines a cone with an open angle equal to the slope angle.

The following diagram (see Figure 8) shows the dimensions of the geometric shapes used to model the rover structure. The maximum stable slope was of 33º with the astronaut and 27º without the astronaut (considered as a mass of 150Kg.).

Figure 8. Maximum stability analysis.

Another alternative to improve the rover overall compactness was to select smaller wheels (Ф 40 cm) instead of the current wheels diameter of 60 cm. diameter. Based on our previous experiences, we have

selected to use bigger wheels for improved mobility in rough terrain for the following reasons:

The maximum height of obstacles will be higher and free space under the chassis will be incremented.

For a given rover velocity the motor speed will be lower and the motor endurance life will be higher.

Resistance to motion in soft terrain will be lower and speed will be higher.

2.3. Electrical Analysis.

The power budget is mainly based on the motor consumption obtaining an estimation of the power of 2kW. Considering as a requirement that the rover must have an autonomy time of 4 h we conclude that the rover requires an energy budget of 8kWh.

It should be noted that the power budget has been computed using some assumptions about the mission scenario (wheel motors running 1h at full power). To keep the power consumption sufficiently low on slopes, the chosen motor type must have a low maximal sustainable torque and a high gear-head ratio.

The peak power estimation is also important in order to estimate the maximum current going through the tether in tethered mode. In a worst case scenario, and only for a few seconds, all wheels could be blocked and ask for up to 1200W each one giving a total amount of 4.800 W (25 A). This worst case scenario implies a total power consumption of 5.920W. Next figure shows the location of the lead-acid batteries (4x120Ah-48V).

Figure 9. EGP-Rover batteries location.

2.4. Crew safety considerations

The rover locomotion system will take care of the crew safety while driving autonomously including the following safety considerations:

Holding bar: The astronaut must handle the horizontal aluminium bar located in front of him in order to avoid falling down from the platform.

Pan&Tilt Vision Unit locking: Pan and Tilt mechanical locking is achieved whenever power supply to this unit is disconnected. The torque

necessary to move the PTU from a static position is bigger than 150 Nm, also in "power off" status.

Back sensors: The two back ultrasound sensors are considered as the emergency stop system: if the astronaut falls down from the rover platform and enters into the back sensor coverage area, the back sensors will detect him and the rover will stop immediately.

Rear wheels protective carter: The rover will include a protective carter over the rear wheels steering mechanism so that the crew on board the EGP cannot be entrapped by the back wheel e.g. foot or leg.

Detection of slopes bigger than 20º: In case of outdoor terrain, it is essential that the Rover does not get stuck / tilt over/ fall because the slope it tries to traverse, climb or descend is too steep w.r.t its capabilities. The RVS system will generate a specific slope map and the IMU unit will produce an average slope value for monitoring purposes.

3. STEREOVISION SYSTEM

In order to fulfill EGP mission requirements (pre-landing, during landing and post-landing of a human crew to a Moon or Mars outpost) the RVS completes the current suite of robotic vision systems (a robot Head Stereo Camera for human-remotely driven operations, and a set of arm cameras for supporting robot operations such as tool manipulation [8]. It has to perform the following tasks:

a) Mapping of environment: Generate a high-resolution 3D description of the environment around the expected Crew landing site by Digital Elevation Model (DEM) generation, navigation & localization in the environment and augmenting the DEM. This includes hazard and slope map generation, absolute localization w.r.t. the lander vehicle, and hazard avoidance during EGP motion.

b) Autonomously following an Astronaut by means of target recognition, localization and tracking

c) Detecting and localizing a spot on the Planetary surface that has been marked by the Astronaut using a light pointing device.

The RVS consists of a colour stereo camera set-up mounted on top of the rover to provide a panoramic stereo view using a pan&tilt unit (PTU). The RVS software solution is implemented as independent agent within the EGP-Rover system, providing a simple command interface (Table 1) and data exchange by file access.

The EGP-Rover geo-localization procedure starts with a rough initial rover position and pose. This initial position is refined by RVS geo-localizing the rover

against artificial (coloured or specially shaped) targets mounted on infrastructure (e.g. the Lander) [9]. Once we have acquired a refined rover position and pose with sufficient accuracy the RVS is requested to perform initial high-resolution mapping resulting in a Digital Elevation Map (DEM) and derived products such as slope map, hazard map and roughness map [5].

Table 1: RVS Command Suite.

Then, the rover starts motion and in parallel the RVS provides near real-time hazard maps (again based on local DEMs and derived products) and it updates the EGP position and pose every few meters by visual comparison of the initial DEM texture with the current image (“surface-relative orientation”). At the end of this vision pipeline the RVS refines the previous existing DEM with later updates resulting in a larger and accumulated DEM version as shown in Figure 10.

Figure 10. DEM (elevations colour-coded) generated from a panorama (left circular part), and subsequently augmented DEM after Rover motion.

Auxiliary functions for astronaut and light spot detection & tracking are executed by a simple solution based on colour spot detection in HSI (Hue-Saturation-Intensity) Space. This capability was demonstrated during Astronaut following tests (see Figure 11).

Figure 11. Astronaut follow-on mission.

During these tests several coloured targets were used (right image from stereo pair) as shown in Figure 12. The wide rectangle indicates the (epipolar) search space determined by pre-search in the left image. A mock-up of the Lander is visible, containing red targets used for RVS absolute localization.

Figure 12. Colour-spot detection by the RVS system.

In the case of the Astronaut follow-on mission the rover motion is synchronized with the vision detection algorithm in a tight integration control loop. In addition, the RVS system also performs a sub-control loop keeping the astronaut silhouette at the centre of image shots.

4. NAVIGATION SUBSYSTEM

4.1. GNC architecture

From a guidance, navigation and control (GNC) point of view the cost-to-go estimate is based on the best knowledge of the world, and the new information could become available beyond the detection range.

The optimal route between two points separated by a certain range generally is to be performed in several iterations since the perfect knowledge of the environment usually is not available to the EGP-Rover GNC (RGNC) system. In each of this iteration, the RGNC computes the optimal route supposing that the unknown area does not have obstacles as depicted in

Figure 13. This way, the RGNC can optimize the collision free path length in the virtual world (known area plus free-obstacle known area) performing sequential iteration, as new region are known.

Figure 13. Path-planning including waypoints and boundaries.

On the other hand, by requirement the rover should be able to overcome steps/rocks of 20 cm, therefore we can consider 20 cm as the minimum value that the Hazard Avoidance system must detect and consider as hazardous objects. Furthermore the detection range provided by the stereo camera based on the Hazard Avoidance function should be based on a trade-off between the length of the mast (in order to reduce the dead-areas), the minimum resolution size of the stereo camera and the maximum velocity of the rover.

The following diagram (see Figure 14) shows the RGNC architecture. The main components are:

Navigation, in charge of providing an estimation of the rover state (position, velocity, orientation and orientation rate) from the sensors measurements and estimates.

Path planning, in charge of finding the best (in terms of safety, suitability and required travel time) path from the current rover location to the target location over the available DEM model.

Guidance, in charge of applying in real-time the pre-defined path planning to the following waypoint and eventual short-range path planning update (due to wheels slip, small obstacles …).

(Translational) Control, in charge of correcting the real rover translational state to the desired rover translational state. The controller function will be designed and developed targeting towards an Ackermann control system included in the Rover Wheels Management function.

Rover Wheels Management, in charge of translating control commands to wheels actuations.

Sensors Pre-processing, in charge of formatting the received sensors measurements and estimates according to GNC needs and to perform some basic pre-processing (e.g. noise filtering).

Figure 14. EGP-Rover RGNC architecture.

4.2. Rover Sensors

Regarding odometry sensors, the EPOS controller is in charge of measuring the motor axis rotations trough dedicated encoders. From these measurements, the displacement of individual wheels can be estimated. An estimation of the combined odometry is computed using the displacement of the front wheels (front left and front right) and assuming no slippage. The resulting displacement is given by Eq. (1):

0 (1)

EGP-Rover odometry computation is based on the average readings from front wheels encoders. Only front wheels are used because they are just traction wheels. Rear wheels are not included in the odometry computation as they are driving wheels and can suffer much more from slippage. Wheels encoders have 3 channels with 500 counts per turn and a maximum operating frequency of 100 KHz.

The EGP-Rover includes a low cost IMU unit as a second navigation sensor. The IMU will allow for: 1) high rate navigation estimation availability, 2) stereovision localization function working gaps (e.g. degraded images, deficient illumination, sky pointing, Sun blinding), 3) high-rate dynamic information for state prediction in hybrid Kalman-type navigation filter and 4) arriving to a safe area (with rough accuracy) in case of total failure of the stereovision.

As main drawback we have that calibration and re-calibration is needed (mandatory) due to the fact that IMU measurements suffer from fast degradation (accuracy is quadratically degraded with time). This system will be considered as complementary with accurate low rate stereo vision system measurements and redundant with respect to mechanical odometers from the locomotion platform. Current chosen IMU

model is the Systron Donner Miniature MEMS Quartz IMU (MMQ50). Using these IMU performances we have performed an analysis reaching a final error position less than 30cm by assuming the following criteria:

Worst case scenario with IMU bias in the same direction than acceleration and initial error position in the direction of the IMU bias.

Constant acceleration of 0.25m/2.

Nominal velocity of 0.75 m/s.

Initial calibration will reduce accelerometer bias by a factor of 10 (from 3 mg to 0.3 mg).

4.3. Sensor data fusion

Estimation of the rover state (position, velocity, orientation and orientation rate) is performed by an Extended-Kalman filter (EKF). The EKF will gather measurements from propioceptive sensors as the mechanical odometers and the IMU and exteroceptive sensors as the stereovision system.

All those measurements will be fused using a tight integration method as shown in next Figure 15. In this filter, the VS information and the IMU mechanization-derived position, velocity and orientation are used to predict VS position and orientation estimates. The VS measurements are then combined with IMU predicted estimated measurements to build a differential observable which can be used to obtain the IMU error estimates and provide the final navigation output.

Figure 15. IMU-RVS tight integration.

4.4. Steering control

The Ackermann model will be adapted to the EGP-Rover by assuming that the virtual wheel is located in the middle point of the front wheels axis as shown in Figure 16. The angle of the virtual wheel is expressed by α and the linear speed by v. Dimensions w and l describe the track width and the track length of the robot. The Centre of Rotation (COR) describes the centre of the rotation of the rover motion.

Figure 16. Ackermann steering control.

To simplify the appearance of the formulas, instead of the curvature K or the curvature radius r, the steering angle of the virtual wheel is used in Eq. (2):

arctan arctan (2)

tan

From this virtual steering angle , the rover rotation speed around its instantaneous centre of rotation is given by Eq. (3):

(3)

4.5. Proximity sensors

The ultrasound system developed in the framework of the EGP-Rover project (see Figure 17) is an independent and standalone system composed by an embedded microprocessor electronic card and 8 ultrasound sensors (6 in the front area and 2 in the back area of the rover) with a 15º field-of-view.

Figure 17. Ultrasound sensors in the rover front area.

The main goal of this system is to be used as a collision avoidance real-time module for obstacles higher than 20 cm and close to the rover within a configurable footprint area (from 20cm to 3m).

After calibration of the sensors, ground objects higher than 20cm will be recognized as obstacles by the ultrasound system and the rover will stop immediately for safety reasons.

4.6. TC/TM

The EGP-Rover provides a full set of telecommands able to command the rover through six different communication channels (system, nominal, get/set, asynchronous and RVS commands). The asynchronous channel is meant for direct locomotion commands at 50 Hz (Ackerman virtual wheel speed, speed and jog or individual wheel steerings).

Figure 18. Screenshot of the telecommands console.

5. CONCLUSION

The EGP-Rover supports various operational modes, such as finding & following a moving Astronaut, carrying a box together with the Astronaut, or finding a specific location on the ground where the Astronaut is pointing using a special light pointing device. Vision and locomotion subsystems therefore support the detection and tracking of an astronaut (again making use of special targets) and the detection of a light beam for spotting an interesting target point.

RVS tests were successful, some additional tests were done to mitigate the varying environment conditions (illumination; albedo & BDRF of simulated Lunar landscape). Due to its modular concept and the simple interface, various updates in terms of future sensors to be used (e.g. a 3D TOF (Time-of-Flight) camera could replace the stereo capability generating DEM’s in real-time) and more enhanced target recognition concepts can be implemented without major effort.

The EGP-Rover, equipped with the two Eurobot Arms, was successfully tested by TAS-I and ESA in a dedicated Moon-like environment at Dutch Space facilities.

The main focus was to evaluate, enhance and prove the abilities of the concept to be a valid candidate solution for supporting future human and robotic space

exploration to Moon, Mars and other Planets (outpost assembly activities are proposed as a promising application and research area for the EGP concept).

In particular two main objectives were fulfilled: a) on-surface autonomous robotic operations in preparation to human arrival, b) on-surface cooperative operation i.e. human to robot joint tasks during human presence.

Further enhancements could be done at the low level locomotion with new techniques of control based on robust model predictive controllers (MPC); also, the increasing level of autonomy (E4 ECSS level as goal-oriented commanding) and integration between all subsystems (robot arms, RVS and RGNC).

Additionally, situational awareness of the astronaut would be increased by adding a specific touch screen controller in his arm and of the rover itself by adding some other exteroceptive sensors like a panoramic imager or hazard cameras. This will allow performing faster operations on one side and on the other side to interact in a more smooth way with the astronaut.

6. REFERENCES

[1] F. Didot et al., Eurobot: Testing the Co-operation between Humans & Robots, IROS 10, 2007.

[2] S. Estable et al., Eurobot Control Station ECoS: The Control Station of the Eurobot Underwater Model; ASTRA Workshop 06, 2006.

[3] F. Didot et al., Eurobot Underwater Model: System Overview, Tests Results & Outlook, i-SAIRAS 02,2008.

[4] A.Medina et al., Astronaut follow-on navigation with a servicing rover, DRHE 2010 (Toulouse, France).

[5] G. Paar et al, “Requirements and Solutions for ExoMars Rover Panoramic Camera 3D Vision Processing”. Geophysical Research Abstracts, Vol. 9, 03901, 2007.

[6] M. Hoepflinger, “Description of the Locomotion Control Architecture on the ExoMars Rover breadboard. “, ASTRA 2008, 10th.

[7] Lanza, P., Estable, S. et al., Eurobot Vision System, an Integrated Approach: Results & Recommendations, iSAIRAS 2008.

[8] E. Zereik, A. Sorbara, G. Casalino and F. Didot, “Force/Vision-Guided Grasping for an Autonomous Dual-Arm Mobile Manipulator Crew Assistant for Space Exploration Missions”, International Conference on Automation Robotics and Control Systems, Orlando, USA, July 2009.

[9] G. Paar, A. Medina, K. Furuya, T.Nunner, A.Bauer, E.Pensavalle, “Computer Vision for a Planetary Service Rover”, EPSC 2010, Vol5, EPSC2010-435.