Planning Robotic Assembly by Reversing Disassembly

This direction is merged into another direction: assembly-by-disassembly. Its last version stays readable and citable here.

Version 1 of 122 papers37 conceptsfrozen 2026-10-04

For a robot, putting a complex object together is significantly harder than taking it apart. In robotics, this is often framed as the "narrow passage" problem: finding the exact trajectory to insert a peg into a hole or a brick into a socket is computationally expensive because the space of successful movements is tiny compared to the space of all possible movements.

However, a recurring insight across decades of research—from early industrial robotics to modern bimanual learning—is that the most efficient way to plan an assembly is to simulate its disassembly. By starting with a completed goal state and figuring out how to remove parts one by one, researchers can generate guaranteed-feasible assembly sequences by simply reversing the clock.

The Computational Ease of Finding the Exit

The core intuition behind this approach is that "finding the way out" is easier than "finding the way in." When a structure is already assembled, the geometric constraints of the parts themselves dictate the only possible paths for movement.

As Bingjie Tang, Iretiayo Akinola, and their colleagues demonstrate in AutoMate, reversing physics-validated disassembly paths allows a system to generate high-quality training data for contact-rich tasks without human demonstrations. Instead of searching blindly for an entry path into a tight-tolerance socket, the system starts from the "solved" state and moves outward. This strategy is also central to SRSA, where Yijie Guo and Abhishek Gupta use automated disassembly to provide a scalable supervisory signal for learning complex skills that would otherwise suffer from sparse rewards.

Disassembly as a Visual Instruction Manual

Beyond just finding trajectories, disassembly reveals the hidden internal logic of an object. For complex structures like LEGO models, the order of operations is not always obvious from the exterior.

In Learning to Build by Building Your Own Instructions, Aaron Walsman and Dieter Fox describe an agent called InstructioNet. To build an unfamiliar object, the agent first disassembles it, recording a sequence of images that act as a step-by-step manual. By reversing this sequence, the agent discovers internal connections that are invisible in the final state. This "self-generated" instruction approach mirrors early work by Masakazu Ejiri and the Hitachi team in An Intelligent Robot with Cognition and Decision-Making Ability, which established that calculating a disassembly sequence is a computationally efficient way to determine the assembly order of complex engineering structures.

Ensuring Stability and Collision Avoidance

Planning in reverse also solves two critical physical problems: structural stability and workspace crowding.

  1. Stability: In Prompt-to-Product, Ruixuan Liu and others use virtual deconstruction to identify a sequence of intermediate states. Because each step of the disassembly must be physically stable, the reversed assembly sequence is guaranteed to never collapse during construction.
  2. Workspace Crowding: In additive manufacturing, such as 3D printing frames, the robot can easily "box itself in" as the structure grows. Caelan Reed Garrett and Tomás Lozano-Pérez show in Scalable and Probabilistically Complete Planning for Robotic Spatial Extrusion that planning backward from the finished structure ensures the robot always has a clear path to exit the workspace, avoiding reachability dead ends.

Handling Irregular and Interlocking Geometries

The "reverse-to-assemble" logic extends even to fractured or interlocking parts where the "correct" fit is highly constrained. Yan Shen and Ruihai Wu apply this to bimanual robots in BiAssemble. By training a model to predict collision-free disassembly trajectories for broken fragments, the robot learns the underlying "topology of escape." Reversing these trajectories allows the robot to fit irregular shards back together without the arms or fragments colliding.

Similarly, for sustainable construction using interlocking "SL blocks," Yuxi Liu uses directional blocking graphs to map the movement freedom of each component. By recursively removing unblocked elements, the system derives a feasible disassembly order that serves as a blueprint for autonomous, adhesive-free construction.

Whether building with LEGOs, 3D printing bridges, or repairing shattered pottery, the research suggests that the most robust way to move forward is to first figure out how to go backward.

Papers in this version

History & citationVersion 1 · 2026-10-04

Merged into another direction · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_planning_robotic_assembly_by_reversing_d_b9dae2d6,
  title = {Planning Robotic Assembly by Reversing Disassembly},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/planning-robotic-assembly-by-reversing-disassembly/rdr_b9dae2d657ad4d4a90c950438d9fb34c},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_planning_robotic_assembly_by_reversing_d_b9dae2d6_v1,
  title = {Planning Robotic Assembly by Reversing Disassembly},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/planning-robotic-assembly-by-reversing-disassembly/rdr_b9dae2d657ad4d4a90c950438d9fb34c/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}