Flying Control for an Intra-Vehicular Free-Flyer

Yuichi Tsumaki and Ikumi MaedaHirosaki University

3 Bunkyo-cho, Hirosaki 036-8561 JAPANtsumaki@cc.hirosaki-u.ac.jp

Abstract

The shortage of human resources on the InternationalSpace Station (ISS) is becoming a serious problem. Inour previous work, a new space robot system, the ”Intra-Vehicular Free-Flyer System” (IVFFS) was proposed. TheIVFFS supports both non-contact and contact tasks duringintra-vehicular activities (IVA). To achieve this goal, we in-troduced a prototype model called ”Space Humming Bird(SHB)” which has a variably structured body to satisfy bothsafety and dexterity requirements. Here we describe severalexperiments involving a prototype model on a planar micro-gravity simulator to confirm the feasibility of our concept.In the experiments, flying control based on a predictive mo-tion display (PMD) is introduced. Point-to-point motions, awall-docking motion, a door-opening task and a detachingmotion are performed.

1 Introduction

Human resources available in orbit have an effect on allactivities in space. However, budget constraints make itdifficult to provide enough astronauts on the internationalspace station (ISS). The use of robot technology can easethis situation. From the beginning, space robots were de-veloped mainly to support extra-vehicular activity (EVA)to free astronauts from dangerous jobs. Recently, severalspace robots have been developed for intra-vehicular activ-ity (IVA). The “Charlotte” intra-vehicular robot led the way[1], but its workspace was restricted to regions close to awall. The personal satellite assistant (PSA) may be one ofthe most sophisticated intra-vehicular robot systems, capa-ble of flying inside the ISS [2]. The PSA is equipped withseveral sensors and cameras, and ground operators can eas-ily communicate with the crew and observe experimentalsetups inside the ISS. However, the PSA can handle onlynon-contact tasks.

In our previous work we have proposed the Intra-Vehicular Free-Flyer System (IVFFS) [3], [4], which has

Figure 1. Overview of Space Humming Bird.

mobility similar to that of the PSA, and can also performmanipulation, so it can perform contact tasks in addition tonon-contact ones. We have introduced an IVFFS prototyperobot named “Space Humming Bird” (SHB). A salient fea-ture of this robot is that its configuration can be changedto adapt its capabilities for various tasks. Fig. 1 shows anoverview of the SHB. Manipulation is achieved via whole-body motions of the SHB, which can operate within a rela-tively wide workspace and with simple hardware. A hand isplaced on the head at the beak location. When in manipula-tion mode, the hand can be extended to the desired length,and can pinch a small object, push a button or twist a switch.To help with these manipulations, a stereo camera system ismounted in the head, producing an eye-like appearance andensuring that the workspace of the end-effector is alwayswithin the camera’s view. To perform prolonged contact-type manipulation, the robot must be attached to the struc-ture of the cabin; in free-flying mode no such manipulationis possible. To achieve the capability, the robot is equippedwith a locking device consisting of a suction disc attached tothe tail. With this device the SHB can fix its tail on a wall,for example. Note that the use of a suction disc is possi-

pret

curp pdp

curv

curv

Figure 2. An example of the PSA.

ble in a pressurized IVA environment, but there the suctiondevice can attach to any flat surface.

In this paper, several experiments are described that wereperformed with a prototype model on a planar micro-gravitysimulator to confirm the feasibility of our concept. For theexperiments, a flying control based on a predictive motiondisplay (PMD) is introduced. The PMD is an operator sup-port system for the teleoperation based on the accelerationcommand [5]. Point-to-point motions, a wall-attaching mo-tion, a door-opening task and a detaching motion were per-formed.

2 Flying Control

2.1 Predictive Motion Display

In our previous work, a PMD (Predictive Motion Dis-play) was introduced to handle manual teleoperation basedon an acceleration command [5]. The PMD is an operatorsupport system that displays the future position of the robotusing computer graphics to reduce the effect of dynamics.In this study, the PMD is used to control flight, as describedin the next subsection. Here, we give a brief overview of thePMD.

The predicted position ppd after tpre seconds is decidedusing the assumption that the robot keeps its current veloc-ity vcur . Therefore, ppd can be written as follows:

ppd = pcur + vcurtpre (1)

where, pcur is the current position. The predicted orienta-tion can be written in the same manner. Fig. 2 shows anexample of the PMD.

Experiments show that the PMD reduces both the taskcompletion time, especially the stopping time, and the col-lision frequency. Through a simple feedback model, it isclear that the PMD creates a damping factor. As a result, thesystem becomes more stable, and operationality improves.

2.2 Flying Control Based on the PMD

The SHB uses a propulsion system based on propellers.Therefore, it can fly using electrical power. However, thefollowing problems should be addressed.

• Disturbance of airflow in the cabin,• Collision with astronauts,• Requirement to limit the maximum velocity for safety,• Suppression of overshoot to avoid collision with walls

It is difficult to control the SHB with feed-forward controlbased on the model, because there are unpredictable distur-bances in the cabin. In addition, travel close to the wallis required to perform contact tasks. Therefore, overshootshould be avoided so as to avoid collisions with the wall.From these requirements, a stable feedback control is intro-duced for point to point (PTP) motion in this study. First,velocity control is employed to restrict maximum velocity,so the thrust of propeller u is defined as follows:

u = kf (vref − vcur). (2)

vref = kp(xref − (xcur + vcurtpre)if vref ≥ vmax

vref = vmax

(3)

where, vref and vcur are reference and current velocities,respectively. vmax is the maximum velocity, xref and xcur

are reference and current positions, respectively. tpre is pre-dictive time and kf , kp are feedback gains. Eq. 3 is basedon the concept of the PMD. Therefore, vcurtpre is relatedto a damping factor [5]. These equations include three pa-rameters kf , kp, tpre. It is easy to find the proper kp by thefollowing procedure. The fastest way to reach a referencepoint is a combination of speed-up toward maximum veloc-ity with maximum acceleration amax, and slow-down withmaximum deceleration. To satisfy such an approach, thereference velocity can be defined as follows:

vref =√

2amaxxcur. (4)

In Fig. 3, the red line shows the reference velocity. How-ever, to retain a safety margin of thrust to compensate forexternal disturbances, the actual reference velocity is set atthe blue line. Then, kp can be defined as the gradient ofthe blue line. Other gains should be decided to suppressovershoot.

3 Experimental Setup

Fig. 4 and 5 show the experimental setup. The prototypeof the SHB is placed in a two-dimensional simulated micro-gravity environment by being mounted on a stone bed, withthe prototype’s horizontal friction reduced to nearly zero

position [mm]

velocity [mm/s]

V_lim

V ref

Figure 3. Reference velocity vs. position.

Figure 4. Experimental setup.

with an air bearing. The prototype has two systems to pro-vide an air bearing: a steel cylinder of high-pressure air, andan electric pump.

Three sets of propellers installed at 120 deg are em-ployed for 3-dof planar motions. Robot localization isachieved by image processing with Hitachi IP7000. A cam-era is installed above the stone bed as shown in Fig. 4, andthe prototype has three red markers to facilitate image pro-cessing.

The prototype has rotational joints at both head and tail.At the head an extensible hand with a ball screw is installed.In addition, a locking device consisting of a suction disk andan air cylinder is mounted at the tail. Another ball screw isutilized for piston actions to vacuum the air-cylinder. In all,four motors power the manipulations.

Fig. 6 shows the control system. The motors are con-

Stereocamera

Hand

Sucking disk

Controlunit

Propeller

Figure 5. SHB prototype.

PC/104

SBC

Wireless LAN

Wireless LAN

Receiver

Video capturecard

Wireless microcamera

RS232C

H8Motor

TITech Driver

Encoder

Control PC

IP7000camera

SHB

Figure 6. Control system.

trolled by an H8 microprocessor connected to a single boardcomputer (SBC). The SBC has a wireless link with a controlPC. Two wireless on-board cameras are installed, yielding afully wireless system with a total mass of 7.2 kg (includingits battery).

4 Docking to the Wall

The SHB must dock on the wall of the cabin to exe-cute a prolonged contact task. Therefore, the prototype isequipped with a locking device consisting of a suction diskand air cylinder. For docking, the suction disk should bepushed with the proper pressure against the wall during ad-hesion. However, the thrust of the propeller is too small forthis, so the following inertia-based procedure is used.

1. Stop at an initial position almost 500 mm from thewall.

Figure 7. Operational interface.

2. Move towards the wall with maximum acceleration, soas to collide with the wall.

3. Start to vacuum-up the air-cylinder just before the col-lision.

All steps are automated. The operator gives a commandthrough the operational interface shown in Fig. 7. The op-erator just click-references a docking point on the wall usingthe upper-left video image. The robot then starts the aboveprocedures automatically. Fig. 8 shows an overview of theprocedure. The upper-right video image shows the resultof image processing for robot localization. The undockingprocedure is also automated.

5 Experiments

5.1 PTP Flying Motion

To confirm the feasibility of flying control, the followingtwo experiments were performed.

• Point-to-point motion with constant orientation.• Point-to-point motion including orientational motion.

A reference position is given by clicking on the video im-age. The reference orientation was initially set to 0 deg inboth cases, with initial orientations set at 0 deg and 90 deg.

Experimental results are shown in Figs. 9 and 10. Fig.9 shows that the SHB reached the reference position with-out over-shoot. It retained a nearly constant orientation and

Click

500 mm

reference point

with maximumacceleration

(1) (2)

(3) (4)

Figure 8. Procedure for attachment.

adhered to the reference velocity without significant errors.Fig. 10 shows that the SHB smoothly achieved both the ref-erence position and orientation. However, a small turbu-lence was found in velocities vx, vy , although angular ve-locity α̇ followed the reference velocity. This is because thecenter of gravity was different from the origin of the bodycoordinates. In addition, the control gain for rotational mo-tion is larger than that for translational motion. As a result,any rotational motion control also produces an effect on thetranslational motion.

5.2 Contact Task

To confirm feasibility of the contact task, followingdemonstrations were performed.

• Automatic docking on the wall.• Door-opening task.• Automatic undocking from the wall.

The door-opening task was performed by manual teleoper-ation with direct vision. Snapshots of the experiments areshown in Fig. 11. In (c) the SHB reaches the initial positionfor docking, then moves toward the wall with maximum ac-celeration. In (e) the SHB docks on the wall. From (f) to(j), the SHB engages its hand with the door lug and opensit. Finally, the SHB unhooks its hand and undocks from thewall.

6 Future Works

Current achievement is still at a basic level. To producea practical SHB, the following requirements must be met.

-100

0

100

200

300

400

500

0 20 40 60 80 100

time [s]

pos

itio

n [

mm

]

XrefXcur

-5

0

5

10

15

20

25

30

35

0 20 40 60 80 100

time [s]

vel

ocit

y [

mm

/s]

Vx_refVx cur

(1) x (4) vx

-100

0

100

200

300

400

500

0 20 40 60 80 100

time [s]

pos

itio

n [

mm

]

YrefYcur

-5

0

5

10

15

20

25

30

35

0 20 40 60 80 100

time [s]

vel

ocit

y [

mm

/s]

Vy_refVy cur

(2) y (5) vy

-60

-40

-20

0

20

40

60

0 20 40 60 80 100

time [s]

att

itu

de

[deg

]

attitude

-6

-4

-2

0

2

4

6

0 20 40 60 80 100

time [s]

an

gu

lar

rate

[d

eg/s

] referenceangular rate

(3) α (6) α̇

Figure 9. Experimental results (translation).

0

100

200

300

400

500

0 20 40 60 80 100

time [s]

pos

itio

n [

mm

]

XrefXcur

-5

0

5

10

15

20

25

30

35

0 20 40 60 80 100

time [s]

vel

ocit

y [

mm

/s]

Vx_refVx cur

(1) x (4) vx

0

100

200

300

400

500

0 20 40 60 80 100

time [s]

pos

itio

n [

mm

]

YrefYcur

-5

0

5

10

15

20

25

30

35

0 20 40 60 80 100

time [s]

vel

ocit

y [

mm

/s]

Vy_refVy cur

(2) y (5) vy

-30

0

30

60

90

120

0 20 40 60 80 100

time [s]

att

itu

de

[deg

]

attitude

-12

-8

-4

0

4

8

0 20 40 60 80 100

time [s]

an

gu

lar

rate

[d

eg/s

] referenceangular rate

(3) α (6) α̇

Figure 10. Experimental results (translationand rotation).

Figure 12. A view of the active window.

• 3D localization• Downsizing of body including propulsion unit• Appropriate operational interface• Automation of fundamental functions• Adaptation to ISS regulations

All of these requirements are common problems for anyintra-vehicular free flyer system. In other words, appropri-ate solutions could be shared to accelerate developments inthe field of IVFFS. Currently, we are planning to utilize theactive window proposed in our previous work for the op-erational interface [6]. The active window enhances robotposture information by 3D motions in the window to aidthe operator’s intuitive assessment, as shown in Fig. 12. Anappropriate operational interface enhances safety and im-proves operational capability.

7 Conclusions

Flying control based on a PMD was introduced for theIVFFS. Experiments showed the feasibility of the proposedmethod in PTP motion control. In addition, sequential mo-tions including wall docking, door-opening and undockingfrom the wall were performed on a micro-gravity simulator,confirming the feasibility of the proposed concept.

References

[1] P. L. Swaim, C. J. Thompson, P. D. Campbell, TheCharlotte (TM) intra-vehicular robot, Proc. of ThirdInt. Symp. on Artificial Intelligence, Robotics, andAutomation for Space, pp. 157–162, 1994.

[2] http://ic.arc.nasa.gov/ic/projects/psa/

[3] Y. Tsumaki, D. N. Nenchev, “Intra-Vehicular Free-Flyer System,” Proc. of the 45th Space Scienceand Technology Conference, pp. 787–790, 2001 (inJapanese).

(a) 0 s (b) 10 s (c) 20 s (d) 30 s

(e) 40 s (f) 50 s (g) 60 s (h) 70 s

(i) 80 s (j) 90 s (k) 100 s (l) 110 s

Figure 11. Snapshots of the experiment.

[4] Y. Tsumaki, M. Yokohama, D. N. Nenchev, In-tra Vehicular Free-Flyer System, Proc. of the 2003IEEE/RSJ Int. Conf. on Intelligent Robots and Sys-tems, pp. 2547–2552, 2003.

[5] Yuichi Tsumaki, Mami Yokohama, Predictive MotionDisplay for Acceleration Based Teleoperation, Proc.of the 2006 IEEE Int. Conf. on Robotics and Automa-tion, pp. 2927–2932, 2006.

[6] Yuichi Tsumaki, Satoshi Kawai, Vincent Hugel,Patrick Bonnin, Pierre Blazevic, An Operational Inter-face Based on the Active Window for Rescue Robot,SI2006, 1G1-6, 2006 (in Japanese).