Yutu-2

NASA/Goddard/Arizona State University. Public domain (NASA / US government work).

Overview

Yutu-2 is the rover element of Chang’e 4 and the first rover to operate on the lunar far side [1], [5].

Specifications

Parameter	Value	Source
Mass	135 kg	[6]
Wheels	6, all driven; front and rear pairs steerable	[3]
Suspension	rocker-bogie, internal differential
Wheel construction	aluminum alloy hub, mesh, 24 grousers in two staggered rows
Slope capability	20 degrees
Obstacle height	200 mm
Working modes	7
Drive segment	4 to 7 m	[1]
Night survival	radioisotope heater units, hibernating with the mast stowed

Mission profile

Parameter	Value	Source
Landing	3 January 2019, Von Karman crater, South Pole-Aitken basin	[1], [5]
Landing point	177.5884 deg E, 45.4565 deg S	[1]
Distance driven, first twelve lunar days	more than 300 m over 73 waypoints
Distance driven, thirteen lunar days	357.695 m
Distance driven by 3 December 2019	more than 400 m, seventeen lunar days
Position at lunar day 41	Longji site, 45.44 deg S, 177.56 deg E	[3]

Site selection

Landing area selection for Chang’e-4 was bounded by three engineering constraints. Thermal control required candidate areas close to 45 degrees S; relay communication capability required longitudes within 161 deg E to 161 deg W, because link performance degrades beyond 180 deg E plus or minus 19 deg; and slope and obstacle limits required an average topographic slope below 8 degrees over a flat area larger than 50 by 30 km [5]. Von Kármán, 186 km across and about 5 km deep in the 2500 km South Pole-Aitken basin, and Chrétien were the two craters meeting all three; Von Kármán’s floor between 176.4 and 178.8 deg E and 45 to 46 deg S was selected, with Chrétien held as backup against a one-day landing delay, which would drift the ground track about 13 degrees in longitude.

Mobility

The rocker-bogie suspension carries an internal differential connecting the left and right bogie assemblies to the vehicle, unlike the Mars Science Laboratory and Mars 2020 arrangement where the differential sits outside the body on a top-deck pivot [3]. All six wheels are driven; only the front and rear pairs steer, with a steering offset between the steering axis and the wheel. Each wheel is an aluminum alloy hub with a mesh tread carrying 24 grousers in two staggered rows, giving a gradeability of 20 degrees and an obstacle height capability of 200 mm [3].

Wheel-terrain behavior at the site was recovered by fitting a grouser-wheel terramechanics model to the 41st lunar day traverse, which was run deliberately close to the locomotion margin on sloping ground [3]. The wheels worked mostly in skid rather than slip, that is at negative slip ratio with wheel circumferential velocity below travel velocity, and sustained lateral slippage throughout. Sweeping internal friction angle from 25 to 55 degrees, predicted endpoint deviation from the planned path stayed within 0.762 m; median slip ratio moved from -0.035 to -0.002 across that range, while side slip angle rose from 0.742 to 1.635 degrees as lateral shear weakened [3].

The inferred regolith parameters at the Chang’e-4 site are an internal friction angle of 21.5 to 42.0 degrees with associated cohesion of 520 to 3154 Pa, and an external friction angle, describing wheel-to-soil roughness in the lateral direction, of 8.3 to 16.5 degrees [3]. The shear characteristics resemble Apollo 12 samples but the cohesion is higher than that measured on most nearside missions. For comparison, in-situ relative density across the Apollo sites is about 65 percent in the top 15 cm and above 90 percent below 30 cm [8].

Power and energy

Two deployable solar panels, one either side of the body, are the primary source, charging lithium-ion batteries. Generation capacity in watts and battery capacity in watt-hours are not published in the peer-reviewed literature, and the mission’s published engineering documentation is not open.

Two power-driven constraints do appear explicitly in the operations literature. Energy availability is one of the named constraints the teleoperation system must plan against, alongside environmental temperature, illumination, measurement control, and relay communications [1]. And because direct sunlight is required to maintain the solar supply, terrain occlusion of the Sun is computed as a routine operational product rather than assumed away.

Thermal

Radioisotope heater units maintain survival temperature through the lunar night, when the panels fold against the body. The rover also hibernates through local noon, when thermal load peaks, so its working periods are the two shoulders of each lunar day. Operating and survival temperature limits are not published in the open literature retrieved here. Thermal control requirements drove the landing latitude to about 45 degrees S [5], where mean surface temperature runs from 350 K at local noon to 89 K before sunrise [7].

Compute and avionics

Processor, memory, and radiation tolerance approach are not published in the open literature. What is documented is the division of labor the avionics support: the rover runs local planning, emergency obstacle avoidance, and emergency protection actions onboard, while route selection and task planning are done on the ground [1].

Autonomy

Yutu-2 uses a combination of ground-based teleoperation and onboard autonomous control, with teleoperation described by its operators as the most important control method [1]. The rover has seven working modes covering different environments and locomotion conditions [3].

The cycle runs as follows. The rover images its surroundings and returns those images with telemetry. The teleoperation center reconstructs the surrounding geography and builds a telepresence representation from them. Scientists select detection targets; the center then performs task planning to fix the work content and driving route, and issues a set of commands uplinked to the rover, which executes them [1]. Onboard autonomy covers local planning within that route, emergency obstacle avoidance, and emergency protection actions.

Before each movement the rover commands its hazard avoidance cameras, navigation cameras and panoramic cameras to capture stereo images at several angles [1]; Pancam is two identical optical systems on the mast [3]. Local digital elevation models at 0.02 m resolution are routinely produced at each waypoint from the Navcam and Hazcam imagery, and those DEMs feed the obstacle recognition method. A 0.03 m resolution digital orthophoto map generated from the descent imagery, covering 211 by 187 m, is one of the base maps for overall path planning [1].

Sixteen pairs of Navcam images at a fixed pitch angle produce the orthophoto at each waypoint, but the mast geometry leaves a blind spot directly beneath the rover that Navcam cannot see. Hazcam stereo fills it, though Hazcam orthophotos are used only within 3.5 m because resolution falls off at range [1]. At one waypoint Hazcam resolved a pit about 45 cm across and 6 cm deep with a stone inside that Navcam had not.

Movements target 4 to 7 m each [1], against a vehicle obstacle-crossing capability of 200 mm and a slope limit of 20 degrees [3]. Obstacle decisions are made against a crossing threshold; the 45 cm pit did not exceed it, and pits of 5 to 6 cm depth were driven past rather than around. Planned three-dimensional paths are projected back onto the Navcam mosaic and Hazcam images for visual verification before execution.

Path planning is posed as two sub-problems, building a cost map of the environment and searching it for an optimal path, subject to rover performance indicators including curvature constraints, obstacle-surmounting capability, and communication conditions [1]. The cost model developed for Chang’e-4 combines non-time-varying factors, slope, roughness, height gradient, and combined obstacles, with time-varying factors, illumination, communication geometry, and shadowed areas. Communication geometry and illumination enter as cost terms rather than as post-hoc checks, which follows from the far-side site, where relay link availability was already an engineering constraint on landing area selection [5].

Terrain occlusion is computed by skyline extraction. The Navcam or Pancam takes a monocular image of the surrounding terrain, and the boundary between sky and surface is extracted by an edge detection method using large-scale unilateral uniform constraints. The absence of a lunar atmosphere helps here: with no diffuse scattering, the sky is dark against a bright surface and the boundary is sharply defined. The extracted skyline is corrected for lens distortion, rover attitude, and mast angle to give skyline altitude and azimuth. The Sun or the relay satellite is then declared occluded whenever its altitude angle falls below the skyline altitude in its direction [1]. This product drives the timing and placement of sleep and reboot commands. A skyline image is taken about every two lunar days near local noon, when solar altitude is highest; six had been made by the end of twelve lunar days. Computed skyline altitude angles were all below 3 degrees, with average error between them under 0.02 degrees, placing the rover about 0.77 degrees clear of the northern mountain and 0.17 degrees clear of the western one, so no distant occlusion constrained the sleep and reboot points [1].

In the Chang’e-4 landing area impact density is high enough that only about 2.5 percent of the surface is relatively flat plain between hazardous craters [1], against a landing-area selection criterion of average slope below 8 degrees [5].

Communications

The far side has no line of sight to Earth. All commands and telemetry route through Queqiao, a relay satellite in a halo orbit about the Earth-Moon L2 point, launched in May 2018 ahead of the lander [2]. The relay is the enabling element of the mission rather than a convenience. It carries a 4.2 m deployable umbrella antenna, links to the lunar surface at X-band, and links to the Earth ground station at S-band [2]. Forward links use a 20 W X-band solid-state amplifier. The rover’s own link is UHF to the lander, which decodes and forwards to the relay.

Rover data rates and window durations are not available: the relay design papers that carry them are behind publisher access controls. Relay communication conditions are treated as a planning constraint at the same level as energy and thermal state [1], and relay visibility is computed per-position from the skyline analysis described above.

Payload and instruments

Lunar Penetrating Radar (LPR). Dual-frequency. Channel 1 is centred at 60 MHz with a range resolution of 1 to 2 m in mare basalt; channel 2 is centred at 500 MHz and penetrates about 35 m [4].

Visible and Near-infrared Imaging Spectrometer (VNIS). Covers 900 to 2395 nm at a default 5 nm sampling interval in the short-wave infrared channel, whose field of view is a circular region of 54 pixel radius in the CMOS image; measurement uncertainty is 5 percent for visible and near-infrared and 7 percent for short-wave infrared [3]. It is used on fresh soil exposed in the rover’s own wheel ruts, on material sputtered from impact craters, and on small rocks [5].

Panoramic camera (Pancam). Two optical systems of identical function on the mast, used for high-resolution topographic mapping and photogrammetric determination of rover position [3].

The engineering cameras are also perception instruments: hazard avoidance cameras, navigation cameras on a mast with pitch and yaw drives, and panoramic cameras [1], [3].

Modes of operation

Drives are short and planned within each lunar day. The rover hibernates through night and through local noon, when thermal load peaks. Sleep and reboot commands are issued at positions and times chosen from the skyline occlusion analysis, so that the rover wakes and sleeps where Sun and relay geometry permit [1].

Ground operations

Teleoperation is conducted from the Beijing Aerospace Control Center [1], with relay passes routed through Queqiao at Earth-Moon L2 [2]. The planning cycle is per-movement rather than per-sol: at each waypoint the rover images, the ground reconstructs terrain and localizes the rover, scientists choose targets, the center plans the route, and commands are uplinked.

Localization runs in two parallel channels [1]. Dead reckoning executes onboard and gives real-time position and attitude. Visual positioning runs at the teleoperation center in near real time and is more accurate. Over traverses of up to 10 m between neighboring waypoints the two agree to within centimeters to decimeters; when the rover slipped badly the disagreement exceeded 1 m, and the visual solution corrected the dead reckoning error [1]. The imaging strategy for visual localization handles the large scale and rotation differences between waypoints by taking a first image at more than 7 m distance with raised pitch at the previous site, then selecting the best image by imaging angle at the two waypoints; features are extracted with ASIFT and the solution obtained by bundle adjustment [1].

Localization is photogrammetric rather than inertial, and Pancam stereo imagery is the data source for both the topographic maps and the position solution [3]. The landing point was fixed by a multi-camera vision positioning method, exploiting the fact that the lander is a known object of known size within about 1.5 m of the camera [1]. Thereafter, DEMs and DOMs are generated at waypoints from Navcam imagery and used for localization, science planning, and path planning. A coarse-to-fine strategy is used to meet the competing demands of timeliness and accuracy under the constraints of the relay transmission link.

Results

The 60 MHz LPR channel resolved reflectors interpreted as ejecta layer boundaries at apparent depths of 51.8 plus or minus 1.1 m, 63.2 plus or minus 1.2 m and 96.2 plus or minus 3.2 m, with further large-scale ejecta structure below 200 m [4]. The 500 MHz channel covers the upper approximately 35 m, resolving fine regolith over coarser ejecta and the regolith-to-basalt interface. Orbital and spaceborne radar had previously resolved mare and bedrock only at apparent depths of 1 to 1.6 km, and nearside maria structure at hundreds of meters [4].

Terramechanic inversion of the 41st lunar day traverse produced the first estimate of the lateral, as opposed to bearing and shear, properties of farside regolith, an external friction angle of 8.3 to 16.5 degrees [3]. Spectral investigation of the target rock at Longji identified iron- and magnesium-rich low-calcium pyroxene and linked it to ejecta from Zhinyu crater at 45.34 deg S, 176.15 deg E, implying an additional source region for local material at the Chang’e-4 site.

Technologies developed

Relay infrastructure converts the far side into a workable operating region: Queqiao’s 4.2 m umbrella antenna, X-band surface link and S-band Earth link [2] made an area that has no direct-to-Earth line of sight commandable at the per-waypoint cadence the rover requires [1].

Skyline extraction from a single monocular image, corrected for lens distortion, rover attitude and mast angle, yields Sun and relay occlusion directly from rover imagery rather than from a terrain model [1]. Digital twin inversion of a driven traverse recovers soil bearing, shear and lateral parameters from vehicle telemetry without a dedicated geotechnical instrument [3], which is the method Yutu-2 used in place of the penetrometers carried by Lunokhod and Surveyor.