The problem the team set out to solve is deceptively simple. Astronauts on long-duration missions spend roughly a third of their precious time on routine maintenance — hauling trash bags, stowing cargo from resupply capsules, and performing housekeeping tasks that, on Earth, would be forgettable. As NASA gears up for sustained lunar operations and eventual Mars transits, crew time becomes the single most expensive and inelastic resource aboard a spacecraft. Yet building a robotic helper that can reliably tie a trash bag, tether a floating tool, or slide a storage drawer in zero gravity has remained fiendishly complex. The physics of low- and zero-gravity manipulation — where friction vanishes, masses drift, and every contact force sends both robot and target tumbling — defeats traditional robotic control approaches that assume steady ground beneath them.

To crack this, NASA Johnson recently assembled the physical iMETRO test bed: full-scale, meticulously detailed interior mockups of next-generation space vehicles and lunar habitats, populated with custom-designed robotic platforms equipped with arms, grippers, and sensor suites. The new iMETRO Dynamic Simulation mirrors this facility down to the sub-millimeter in a virtual environment, capturing the multibody dynamics, contact mechanics, and momentum-conserving behaviors that define the real hardware’s motion in zero-g. Beneath the hood, the simulator integrates a modern physics engine that solves forward and inverse dynamics for floating-base robots interacting with deformable objects — think of a robot arm pushing against a half-full cargo bag that simultaneously drifts and deforms. The fidelity allows algorithms to experience the cascading perturbations that happen when a gripper makes off-center contact with a free-floating payload, a scenario that has historically caused trained policies to fail catastrophically once transferred out of simplified simulators.

The team’s paper, presented by first author Nikki Hart, a Rice doctoral student and NASA Pathways intern, showcases an arresting result: using the simulation, the researchers developed a new software application for remote robot operation and successfully deployed it on the physical iMETRO hardware at Johnson Space Center with full functionality in under a single working day. “We had it fully operational in less than a day,” Hart said, emphasizing that this kind of end-to-end transfer has traditionally required weeks of on-site tuning. The achievement relies on the simulator’s ability to faithfully model not just gross robot motions but also sensor latencies, actuator bandwidth limits, and the idiosyncratic communication delays between a remote command station and the physical test bed — barriers that often derail pure software-to-hardware handoffs.

Lydia Kavraki, University Professor at Rice and a senior figure in algorithmic robotics, described the simulator as a bridge that makes space robotics research “accessible to the global robotics community.” By releasing the entire framework as open-source, the joint team has effectively given away a digital copy of NASA’s multimillion-dollar test facility. Academic labs in Nairobi, start-ups in São Paulo, and hobbyist groups in Helsinki can now prototype, validate, and iterate on control algorithms inside the same virtual environment that NASA uses, then, when ready, remotely schedule a slot on the physical robot to benchmark performance. This model circumvents the historical bottleneck in which only well-funded labs with their own custom zero-gravity simulation rigs could contribute to the field.

The digital twin concept at the heart of iMETRO is aggressively thorough. It does not merely replay prerecorded sensor data or approximate gravity as a constant downward vector; it actively solves the coupled equations of motion for the robot’s articulated body and any objects it touches, accounting for the inertial properties of each link, joint friction, and motor torque saturation. When a simulated robot grasps a floating cylinder, the contact model computes normal and frictional forces at the interface, updating the linear and angular momentum of both bodies in real time. This is essential because even a minor miscalculation in angular momentum can send a grasped object into an uncontrollable spin that the real station crew would have to correct. The simulator also exposes a suite of standardized benchmarking scenarios — such as bimanual bag sealing, drawer manipulation, and grappling a tumbling mock resupply container — that provide a common metric to compare approaches from different teams worldwide.

With the platform now publicly available and documentation hosted alongside the open-source code, the team expects a rapid proliferation of novel methods. Early adopters are already testing reinforcement learning agents that discover opportunistic bracing strategies — using a wall or handrail to stabilize motion — as well as model-predictive controllers that plan sequences of gentle nudges rather than firm grasps, a tactic uniquely suited to the momentum-conserving physics of space. Because the simulation can be accelerated to run many times faster than real time, training a policy for a task like clearing a logistics module might take hours instead of the weeks required in a physical lab.

The implications stretch beyond the International Space Station’s successors. The same simulation infrastructure can be retargeted to model lunar lava-tube habitats or the interior of a Mars transit vehicle, simply by importing new geometry files and adjusting the gravity vector — or setting it to zero. For the Artemis generation, where astronauts will live and work in orbit and on the surface in a continuous cycle, robotic aides that can seamlessly transition from zero-g to one-sixth-g partial gravity will be force multipliers. By open-sourcing the simulation stack now, NASA and Rice are laying the groundwork for an ecosystem in which algorithmic breakthroughs can move directly from a research paper into a flight-qualified demonstration without a multiyear, bespoke integration phase. As Kavraki put it, this new virtual testbed “significantly advances research and development in intravehicular space robotics,” and the open invitation to the world’s roboticists is already generating a groundswell of experimental pull requests. For astronauts of the 2030s, the gift of an extra hour a day to do science rather than janitorial work may start with a humble line of simulation code pushed from a laptop on Earth.

Subject of Research: Open-source dynamic simulation environment for intravehicular space robotics
Article Title: The iMETRO Dynamic Simulation: An Open-Source Simulator for Intravehicular Space Robotics Research
News Publication Date: 2026 (presented at ICRA 2026, Vienna)
Web References: NASA Technical Reports Server paper link; Lydia Kavraki profile; Rice University School of Engineering and Computing
References: N. Hart et al., “The iMETRO Dynamic Simulation: An Open-Source Simulator for Intravehicular Space Robotics Research,” presented at IEEE ICRA 2026, NASA/TM-20260001395
Image Credits: (Not provided in source material)

Keywords

space robotics, digital twin, open-source simulation, intravehicular robotics, NASA, iMETRO, zero-gravity manipulation, dynamic simulation, robotic maintenance, remote robotics testbed