RSO Inspection

This example demonstrates the configuration of a resident space object (RSO) inspection environment, in which a servicer spacecraft circumnavigates a RSO to image the illuminated facets.

[1]:
from functools import partial
from importlib.metadata import version
from typing import ClassVar

import numpy as np
from Basilisk.architecture import bskLogging
from Basilisk.utilities.orbitalMotion import elem2rv
from Basilisk.utilities.RigidBodyKinematics import C2MRP

from bsk_rl import ConstellationTasking, act, data, obs, sats, scene
from bsk_rl.obs.relative_observations import rso_imaged_regions
from bsk_rl.sim import dyn, fsw
from bsk_rl.utils.orbital import (
    fibonacci_sphere,
    random_orbit,
    random_unit_vector,
    relative_to_chief,
)

bskLogging.setDefaultLogLevel(bskLogging.BSK_WARNING)

RLlib is actively developed and can change significantly from version to version. For this script, the following version is used:

[2]:
version("ray")  # Parent package of RLlib
[2]:
'2.35.0'

Defining the Satellites

First, the RSO satellite is configured. A simple model is used that has no actuators modelled and just deterministically points nadir.

[3]:
class RSOSat(sats.Satellite):
    observation_spec: ClassVar[list[obs.Observation]] = [
        obs.SatProperties(dict(prop="one", fn=lambda _: 1.0)),
    ]
    action_spec: ClassVar[list[act.Action]] = [act.NadirPoint(duration=1e9)]
    dyn_type = (dyn.ConjunctionDynModel, dyn.RSODynModel)
    fsw_type = fsw.FSWModel

Arguments for the satellite are configured for smooth pointing behavior.

[4]:
rso_sat_args = dict(conjunction_radius=2.0)

The inspector satellite has a more complex configuration. First, an observation function for the sun vector is defined.

[5]:
def sun_hat_chief(self, other):
    r_SN_N = (
        self.simulator.world.gravFactory.spiceObject.planetStateOutMsgs[
            self.simulator.world.sun_index
        ]
        .read()
        .PositionVector
    )
    r_BN_N = self.dynamics.r_BN_N
    r_SN_N = np.array(r_SN_N)
    r_SB_N = r_SN_N - r_BN_N
    r_SB_N_hat = r_SB_N / np.linalg.norm(r_SB_N)
    HN = other.dynamics.HN
    return HN @ r_SB_N_hat

The inspector satellite is configured with observations relating to the relative state and the mission objectives. The satellite is given an action for impulsively thrusting and drifting. The dynamics and flight software models introduce a maximum range check, collision checking orbital maneuvers, and RSO inspection.

[6]:
class InspectorSat(sats.Satellite):
    observation_spec: ClassVar[list[obs.Observation]] = [
        obs.SatProperties(
            dict(prop="dv_available", norm=10),
            dict(prop="inclination", norm=np.pi),
            dict(prop="eccentricity", norm=0.1),
            dict(prop="semi_major_axis", norm=7000),
            dict(prop="ascending_node", norm=2 * np.pi),
            dict(prop="argument_of_periapsis", norm=2 * np.pi),
            dict(prop="true_anomaly", norm=2 * np.pi),
            dict(prop="beta_angle", norm=np.pi),
        ),
        obs.ResourceRewardWeight(),
        obs.RelativeProperties(
            dict(prop="r_DC_Hc", norm=500),
            dict(prop="v_DC_Hc", norm=5),
            dict(
                prop="rso_imaged_regions",
                fn=partial(
                    rso_imaged_regions,
                    region_centers=fibonacci_sphere(15),
                    frame="chief_hill",
                ),
            ),
            dict(prop="sun_hat_Hc", fn=sun_hat_chief),
            chief_name="RSO",
        ),
        obs.Eclipse(norm=5700),
        obs.Time(),
    ]
    action_spec: ClassVar[list[act.Action]] = [
        act.ImpulsiveThrustHill(
            chief_name="RSO",
            max_dv=1.0,
            max_drift_duration=5700.0 * 2,
            fsw_action="action_inspect_rso",
        )
    ]
    dyn_type = (dyn.MaxRangeDynModel, dyn.ConjunctionDynModel, dyn.RSOInspectorDynModel)
    fsw_type = (
        fsw.SteeringFSWModel,
        fsw.MagicOrbitalManeuverFSWModel,
        fsw.RSOInspectorFSWModel,
    )

Generous configurations are used for the inspector, allowing for “sloppy” attitude control with a low simulation step rate.

[7]:
inspector_sat_args = dict(
    imageAttErrorRequirement=1.0,
    imageRateErrorRequirement=None,
    instrumentBaudRate=1,
    dataStorageCapacity=1e6,
    batteryStorageCapacity=1e9,
    storedCharge_Init=1e9,
    conjunction_radius=2.0,
    dv_available_init=10.0,
    max_range_radius=1000,
    chief_name="RSO",
    u_max=1.0,
)

Environment Generation

A satellite argument randomizer is defined to configure the initial state of the satellites. The RSO is put into a random orbit with an apogee and perigee between 500 km and 1100 km. The inspector is placed in the region 250 to 750 meters from the RSO, with up to 1 m/s of relative velocity. Finally, the RSO’s attitude and body rate are set up to be in the nadir-pointing initial configuration.

[8]:
def sat_arg_randomizer(satellites):
    # Generate the RSO orbit
    R_E = 6371.0  # km
    a = R_E + np.random.uniform(500, 1100)
    e = np.random.uniform(0.0, min(1 - (R_E + 500) / a, (R_E + 1100) / a - 1))
    chief_orbit = random_orbit(a=a, e=e)

    inspectors = [sat for sat in satellites if "Inspector" in sat.name]
    rso = next(satellite for satellite in satellites if satellite.name == "RSO")

    # Generate the inspector initial states.
    args = {}
    for inspector in inspectors:
        relative_randomizer = relative_to_chief(
            chief_name="RSO",
            chief_orbit=chief_orbit,
            deputy_relative_state={
                inspector.name: lambda: np.concatenate(
                    (
                        random_unit_vector() * np.random.uniform(250, 750),
                        random_unit_vector() * np.random.uniform(0, 1.0),
                    )
                ),
            },
        )
        args.update(relative_randomizer([rso, inspector]))

    # Align RSO Hill frame for initial nadir pointing
    mu = rso.sat_args_generator["mu"]
    r_N, v_N = elem2rv(mu, args[rso]["oe"])

    r_hat = r_N / np.linalg.norm(r_N)
    v_hat = v_N / np.linalg.norm(v_N)
    x = r_hat
    z = np.cross(r_hat, v_hat)
    z = z / np.linalg.norm(z)
    y = np.cross(z, x)
    HN = np.array([x, y, z])
    BH = np.eye(3)

    a = chief_orbit.a
    T = np.sqrt(a**3 / mu) * 2 * np.pi
    omega_BN_N = z * 2 * np.pi / T

    args[rso]["sigma_init"] = C2MRP(BH @ HN)
    args[rso]["omega_init"] = BH @ HN @ omega_BN_N

    return args

The scenario is configured to set the RSO geometry as a sphere with 100 points at a radius of 1 meter. Points must be imaged within 30 degrees of their normal, with illumination coming from no more than 60 degrees from normal. The inspector must be within 250 meters to inspect the RSO.

[9]:
scenario = scene.SphericalRSO(
    n_points=100,
    radius=1.0,
    theta_max=np.radians(30),
    range_max=250,
    theta_solar_max=np.radians(60),
)

This scenario uses two rewarders. For the RSO inspection component of the task, a bonus of 1.0 is yielded once at least 90% of the illuminated points have been inspected. The ResourceReward is used to penalize fuel use, with some basic logic add to only apply the reward to the Inspector.

[10]:
rewarders = (
    data.RSOInspectionReward(
        completion_bonus=1.0,
        completion_threshold=0.90,
    ),
    data.ResourceReward(
        resource_fn=lambda sat: sat.fsw.dv_available
        if isinstance(sat.fsw, fsw.MagicOrbitalManeuverFSWModel)
        else 0.0,
        reward_weight=np.random.uniform(0.0, 0.5),
    ),
)

With all the components defined, the environment can be instantiated.

[11]:
env = ConstellationTasking(
    satellites=[
        RSOSat("RSO", sat_args=rso_sat_args),
        InspectorSat("Inspector", sat_args=inspector_sat_args, obs_type=dict),
    ],
    sat_arg_randomizer=sat_arg_randomizer,
    scenario=scenario,
    rewarder=rewarders,
    time_limit=60000,
    sim_rate=5.0,
    log_level="INFO",
)

Environment Interaction

The environment is reset and randomly stepped through.

Future Work: This example will be updated with an actual trained policy in the future.

[12]:
env.reset()
for i in range(4):
    env.step(dict(RSO=0, Inspector=env.action_space("Inspector").sample()))
2026-09-02 14:53:15,629 gym                            INFO       Resetting environment with seed=565584306
2026-09-02 14:53:15,697 gym                            INFO       <0.00> Environment reset
/opt/hostedtoolcache/Python/3.11.16/x64/lib/python3.11/site-packages/gymnasium/spaces/box.py:130: UserWarning: WARN: Box bound precision lowered by casting to float32
  gym.logger.warn(f"Box bound precision lowered by casting to {self.dtype}")
2026-09-02 14:53:15,699 gym                            INFO       <0.00> === STARTING STEP ===
2026-09-02 14:53:15,699 sats.satellite.RSO             INFO       <0.00> RSO: action_nadir_point tasked for 1000000000.0 seconds
2026-09-02 14:53:15,700 sats.satellite.RSO             INFO       <0.00> RSO: setting timed terminal event at 1000000000.0
2026-09-02 14:53:15,701 sats.satellite.Inspector       INFO       <0.00> Inspector: Thrust clamped from 1.2485895346410074 m/s to 1.0 m/s.
2026-09-02 14:53:15,701 sats.satellite.Inspector       INFO       <0.00> Inspector: Thrusting with inertial dV [ 0.75075612 -0.65674687 -0.07105486] with 3781.12890625 second drift.
2026-09-02 14:53:15,703 sats.satellite.Inspector       INFO       <0.00> Inspector: setting timed terminal event at 3781.1
2026-09-02 14:53:15,703 sats.satellite.Inspector       INFO       <0.00> Inspector: FSW action action_inspect_rso activated.
2026-09-02 14:53:15,734 sats.satellite.Inspector       INFO       <675.00> Inspector: Exceeded maximum range of 1000 m from RSO
2026-09-02 14:53:15,738 data.rso_inspection            INFO       <675.00> Inspected/Illuminated/Total: 0/34/100
2026-09-02 14:53:15,739 data.composition               INFO       <675.00> ResourceReward reward: {'Inspector': np.float64(-0.10761720779062889)}
2026-09-02 14:53:15,739 data.base                      INFO       <675.00> Total reward: {'Inspector': np.float64(-0.10761720779062889)}
2026-09-02 14:53:15,740 sats.satellite.Inspector       WARNING    <675.00> Inspector: failed range_valid check
2026-09-02 14:53:15,743 gym                            INFO       <675.00> Step reward: {'Inspector': np.float64(-1.107617207790629)}
2026-09-02 14:53:15,743 gym                            INFO       <675.00> Episode terminated: ['Inspector']
2026-09-02 14:53:15,744 gym                            INFO       <675.00> === STARTING STEP ===
2026-09-02 14:53:15,744 sats.satellite.RSO             INFO       <675.00> RSO: action_nadir_point tasked for 1000000000.0 seconds
2026-09-02 14:53:15,745 sats.satellite.RSO             INFO       <675.00> RSO: setting timed terminal event at 1000000675.0
2026-09-02 14:53:15,746 sats.satellite.Inspector       INFO       <675.00> Inspector: Thrusting with inertial dV [-0.2867725  -0.0745726   0.68105478] with 5483.86474609375 second drift.
2026-09-02 14:53:15,747 sats.satellite.Inspector       INFO       <675.00> Inspector: setting timed terminal event at 6158.9
2026-09-02 14:53:15,748 sats.satellite.Inspector       INFO       <675.00> Inspector: FSW action action_inspect_rso activated.
2026-09-02 14:53:16,011 sats.satellite.Inspector       INFO       <6160.00> Inspector: timed termination at 6158.9
2026-09-02 14:53:16,024 data.rso_inspection            INFO       <6160.00> Inspected/Illuminated/Total: 0/72/100
2026-09-02 14:53:16,025 data.composition               INFO       <6160.00> ResourceReward reward: {'Inspector': np.float64(-0.07992960867372398)}
2026-09-02 14:53:16,025 data.base                      INFO       <6160.00> Total reward: {'Inspector': np.float64(-0.07992960867372398)}
2026-09-02 14:53:16,026 sats.satellite.Inspector       INFO       <6160.00> Inspector: Satellite Inspector requires retasking
2026-09-02 14:53:16,027 gym                            INFO       <6160.00> Step reward: {}
2026-09-02 14:53:16,028 gym                            INFO       <6160.00> === STARTING STEP ===
2026-09-02 14:53:16,029 sats.satellite.RSO             INFO       <6160.00> RSO: action_nadir_point tasked for 1000000000.0 seconds
2026-09-02 14:53:16,029 sats.satellite.RSO             INFO       <6160.00> RSO: setting timed terminal event at 1000006160.0
2026-09-02 14:53:16,030 sats.satellite.Inspector       INFO       <6160.00> Inspector: Thrusting with inertial dV [-0.36555087 -0.48847812 -0.40758844] with 8420.3720703125 second drift.
2026-09-02 14:53:16,031 sats.satellite.Inspector       INFO       <6160.00> Inspector: setting timed terminal event at 14580.4
2026-09-02 14:53:16,032 sats.satellite.Inspector       INFO       <6160.00> Inspector: FSW action action_inspect_rso activated.
2026-09-02 14:53:16,381 sats.satellite.Inspector       INFO       <14585.00> Inspector: timed termination at 14580.4
2026-09-02 14:53:16,403 data.rso_inspection            INFO       <14585.00> Inspected/Illuminated/Total: 0/72/100
2026-09-02 14:53:16,403 data.composition               INFO       <14585.00> ResourceReward reward: {'Inspector': np.float64(-0.07896248210330115)}
2026-09-02 14:53:16,404 data.base                      INFO       <14585.00> Total reward: {'Inspector': np.float64(-0.07896248210330115)}
2026-09-02 14:53:16,404 sats.satellite.Inspector       INFO       <14585.00> Inspector: Satellite Inspector requires retasking
2026-09-02 14:53:16,406 gym                            INFO       <14585.00> Step reward: {}
2026-09-02 14:53:16,407 gym                            INFO       <14585.00> === STARTING STEP ===
2026-09-02 14:53:16,407 sats.satellite.RSO             INFO       <14585.00> RSO: action_nadir_point tasked for 1000000000.0 seconds
2026-09-02 14:53:16,408 sats.satellite.RSO             INFO       <14585.00> RSO: setting timed terminal event at 1000014585.0
2026-09-02 14:53:16,409 sats.satellite.Inspector       INFO       <14585.00> Inspector: Thrust clamped from 1.3327595064905846 m/s to 1.0 m/s.
2026-09-02 14:53:16,409 sats.satellite.Inspector       INFO       <14585.00> Inspector: Thrusting with inertial dV [-0.28279537 -0.92021845 -0.27060076] with 6429.45263671875 second drift.
2026-09-02 14:53:16,411 sats.satellite.Inspector       INFO       <14585.00> Inspector: setting timed terminal event at 21014.5
2026-09-02 14:53:16,412 sats.satellite.Inspector       INFO       <14585.00> Inspector: FSW action action_inspect_rso activated.
2026-09-02 14:53:16,656 sats.satellite.Inspector       INFO       <21015.00> Inspector: timed termination at 21014.5
2026-09-02 14:53:16,676 data.rso_inspection            INFO       <21015.00> Inspected/Illuminated/Total: 0/72/100
2026-09-02 14:53:16,676 data.composition               INFO       <21015.00> ResourceReward reward: {'Inspector': np.float64(-0.10761720779062889)}
2026-09-02 14:53:16,677 data.base                      INFO       <21015.00> Total reward: {'Inspector': np.float64(-0.10761720779062889)}
2026-09-02 14:53:16,677 sats.satellite.Inspector       INFO       <21015.00> Inspector: Satellite Inspector requires retasking
2026-09-02 14:53:16,678 gym                            INFO       <21015.00> Step reward: {}