ADRAS-J

Astroscale.

Overview

ADRAS-J is a debris inspection spacecraft built by Astroscale Japan for Phase I of JAXA’s Commercial Removal of Debris Demonstration project (CRD2) [6], [21]. Its task was to rendezvous with a large, uncontrolled rocket stage, hold position relative to it, and characterize its condition and motion by imaging. The target has no navigation aid, no grapple fixture and no GPS receiver, so relative navigation had to come from onboard sensing of the object itself [6].

A cooperative client carries markers, a docking interface or a transponder, and a servicer can be designed around them. An unprepared client has none of these and its rotation state is not known in advance. The servicer therefore needs onboard authority to abort an approach [3], [6].

The spacecraft has a mass of 150 kg and was released by a Rocket Lab Electron from Launch Complex 1 in New Zealand on 18 February 2024, into an orbit of about 600 km [2].

Design space for debris approach

Three families of approach to a non-cooperative client appear in the corpus, and they differ mainly in what has flown.

Inspection without contact is what ADRAS-J flew: angles-only navigation at long range, model-based relative navigation closer in, a fixed-point hold and a fly-around, with no capture hardware [6], [21]. It is the first rung of a ladder in which each step is reversible.

Capture demonstration against a cooperative client is what Astroscale’s earlier ELSA-d flew in low Earth orbit. A 175 kg servicer and a 17 kg client with a ferromagnetic docking plate and fiducial pattern were launched in March 2021 to run capture, tumbling-client capture and lose-and-recapture demonstrations, with no servicer-client communication during the demonstrations [7]. A propulsion anomaly in January 2022 disabled four of the servicer’s eight 1 N thrusters and stopped the planned 25 January capture [9]. Before that attempt was abandoned the servicer held 30 m separation for more than seven hours under autonomous station keeping [8]. It later closed from 1700 km to about 159 m of the client on 7 April 2022, handing navigation from the ground to the onboard system [9]. The client was a purpose-built target, which ADRAS-J’s was not [6].

Capture of an existing object has not flown. ADRIOS ClearSpace-1 retargeted from the VESPA adapter to PROBA-1 (94 kg, under 1 cubic meter, no appendages) and plans angles-only far-range navigation, a fly-around down to 50 to 100 m, and hops from a formation-keeping point at about 150 m to a final approach point, with capture fully autonomous [12]. Launch was expected at the end of 2028, so every performance statement there is design intent [12]. The close-proximity chain for the earlier ClearSpace-1 target was validated in simulation and on a hardware-in-the-loop bench with a target mockup, with abort maneuvers leaving the servicer on passively safe drifting trajectories [11]. Net, harpoon, robotic arm and dry-adhesive capture methods are surveyed in a 2025 review, which reports that most have never flown and that its net capture results are simulation only [13].

Angles-only far-range navigation to a passive client has a flight precedent. In the ARGON experiment on PRISMA in April 2012, ground-in-the-loop image processing and batch orbit determination took the Mango servicer from 30 km to a 3 km hold point relative to the noncooperative Tango in about four days, using 1 m/s of delta-v [15]. Eight of ten orbit determination runs held the relative semimajor axis error to 3 m and the mean along-track error to 400 m, about 1 percent of separation, against GPS reference [15]. The processing ran on the ground after each orbit, not onboard in real time [16]. The close-range chain was exercised on the EPOS 2.0 robotic testbed, which closed from 20 to 3 m with monocular pose estimation [16].

Capture of a tumbling client has been tested on the ground only. A laboratory run with two manipulators, one driving a satellite mockup and one carrying a capture hand, converged an extended Kalman filter on pose, velocities and inertia parameters in 85 s and planned a 36.5 s interception, with no microgravity dynamics in the loop [19]. The DLR OOS-SIM facility reproduces servicer-client contact dynamics with two industrial robots in a 23.28 cubic meter common workspace, with a hardware-in-the-loop position error of 0.005 m against its own offline simulation and no flight comparison [20]. A CNES review written in 2012 identified non-cooperative tumbling-target rendezvous and capture as the undemonstrated functions of active debris removal [18]. A 2016 review of capture and removal methods states that no debris had been removed as of 2015 and groups targets into four non-cooperativeness categories mapped to candidate capture methods [17].

Onboard avoidance planning is a separate research line. A two-stage temporal planner that schedules routine observation tasks around debris avoidance, with an orbital transfer assumed as the avoidance decision, was evaluated in six simulated scenarios only, with no flight or hardware test [14].

The ADRAS-J sequence follows the same staging by navigation method, starting from angles-only far-range navigation [6], [12].

Mission profile

Event	Date	Source
Launch on Electron from Launch Complex 1, New Zealand	2024-02-18	[2]
Angles-only navigation phase started	2024-04-09	[21]
Fixed-point observation at about 50 m	2024-05-23	[21]
Fly-around observations	2024-07-15 and 2024-07-16	[4], [21]
First final approach attempt, about 20 m	2024-07-17	[21]
Third fly-around observation	2024-08-13	[21]
Approach to about 15 m, terminated by autonomous abort	2024-11-30	[3], [21]
De-orbit operations; CRD2 Phase I complete	2026-03-25	[21]

The target is a Japanese H-IIA upper stage launched in 2009, about 11 m long, 4 m in diameter and about 3 t [4], [6]. Its orbit is 533 by 597 km at 98.20 degrees inclination [5]. CRD2 Phase II assigns docking with the same upper stage and removal from orbit to a subsequent mission [6].

Specifications

Parameter	Value	Source
Mass	150 kg	[2]
Thrusters	12, 1 N each	[5]
Propellant	ammonium dinitramide monopropellant	[5]
Relative navigation modes	angles-only navigation, model matching navigation	[21]
Collision avoidance	onboard, with autonomous abort authority	[3]

The spacecraft carries no manipulator. Capture hardware belongs to the Phase II vehicle, for which the Payload Attach Fitting on the target is the planned capture point [3].

Approach and inspection sequence

Operations were staged by navigation method more than by distance. Absolute navigation brought the spacecraft near the target, relative navigation with onboard instruments detected and closed on it, and diagnosis and fly-around followed [6]. Angles-only navigation, which uses bearing measurements alone before the target resolves in the sensor, began at a range of several hundred kilometers and handed over to model matching navigation, which fits sensor returns against a model of the object [21]. The company names the algorithm but publishes no sensor specification or navigation accuracy [21].

The fly-around observations of 15 and 16 July 2024 held a controlled fixed-point relative position of approximately 50 m while imaging the stage from varying angles and lighting [4]. The operator reports that its fault detection, isolation and recovery function had already aborted the first fly-around observation attempt in July [3]. Three successful fly-arounds are reported in all [21].

The 15 m approach

On 30 November 2024 the spacecraft closed from a 50 m hold point along a straight-line approach to approximately 15 m below the Payload Attach Fitting, matching relative speed, distance and attitude [3]. The onboard collision avoidance system then triggered an autonomous abort in response to an unexpected relative attitude of the upper stage, and the spacecraft departed as designed without reaching the Capture Initiation Point [3]. The anomaly was on the target, which is the hazard an unprepared client presents. The operator had not described its nature when it published [3].

Sensing and image products

The company lists no sensor performance figures for ADRAS-J [21]. A catalog entry lists the payload as a space debris monitor [5].

The released fly-around video has been used outside the mission. A monocular pipeline of background segmentation, structure from motion and neural implicit surface reconstruction registered all 419 extracted frames, with 16,505 sparse points, a mean reprojection error of 0.410 px and a recovered focal length of 2396.08 px [1]. No CAD model of the stage is public, so the reconstruction is judged only qualitatively and its geometric accuracy is unknown [1].

Pose and motion estimation of an unknown spinning target from stereo imagery has a flown precedent on the ISS. A factor-graph method estimated an unknown SPHERES target’s pose, velocity, center of mass and inertia from ISS data, with position agreeing with the ultrasound metrology to 1.59 cm along the range axis [10]. It ran offline on the ground with the inspector held stationary [10].

Autonomy

Operations combined onboard functions with ground control [6]. The abort on 30 November 2024 was commanded by the spacecraft, not from the ground [3]. For comparison, ELSA-d handed navigation from ground to onboard control only in April 2022 [9].

Propulsion and power

Twelve 1 N thrusters on ammonium dinitramide monopropellant provide translation and attitude control [5]. Solar wings supply power. No power or energy storage figures are published in any source held.

Communications

No band or data rate is given in any source held. The mission released inspection imagery as video [1].

What is not established

Every ADRAS-J performance figure traces to the operator or to a catalog entry copying it: the 15 m and 50 m ranges are approximate and self-reported, with no position-keeping accuracy, dispersion or telemetry [3], [4], [21]. The cause of the November 2024 attitude anomaly is not described in any source held, nor is the target’s tumble state [3], [4]. Sensor specifications and relative navigation accuracy are unpublished [21]. The reconstruction results from the released video carry no ground truth [1]. Capture of an unprepared object has not flown, and the ClearSpace-1, net-capture and avoidance-planning results are simulation or design intent [11], [12], [13], [14]. The ELSA-d anomaly cause is also unexplained [8], [9].