Full-Tilt Model of Motion Planning

• Full-Tilt Model of Motion Planning is a framework where collections of rigid polyominoes in a finite grid are moved collectively by uniform tilt commands.
• It enables universal self-assembly and simulation of computation by using maximal, parallel motion operations and bonding mechanisms.
• The model highlights computational complexity in motion planning through PSPACE-hard problems, addressed via tailored heuristic and sampling-based algorithms.

The full-tilt model of motion planning is a theoretical and algorithmic framework in which collections of rigid polyominoes (most commonly, unit-square tiles) in a finite grid world are actuated by means of uniform, global external forces—so-called "tilts"—that move all subassemblies maximally in a specified direction. This model, originated to describe programmable self-assembly, programmable matter, and algorithmic swarm robotics, has grown into a central testbed for universal construction, computational complexity in motion planning, and algorithmic design for collective particle systems subject to uniform control inputs (Avila-Jimenez et al., 6 Dec 2025, Blumenberg et al., 2023, Balanza-Martinez et al., 2019). The model is equally prevalent in the analysis of hardware swarms driven by uniform fields and in universal computation by mechanical means.

1. Formal Model Definition and Tilt Operations

Let $B \subset \mathbb{Z}^2$ be a finite rectangular grid partitioned into open positions and blocked "wall" cells. Rigid objects—polyominoes, which are finite, connected sets of occupied open cells—may be placed on $B$, subject to non-overlap and wall exclusion. Of key interest are special cases: singleton tiles (size 1) and dominoes (size 2, i.e., $1\times 2$ or $2\times 1$).

A configuration $C = (B, P)$ is a board $B$ with a set (or multiset) $P$ of non-overlapping polyominoes occupying only open cells.

Given a direction $$d \in \{\mathsf{N}, \mathsf{E}, \mathsf{S}, \mathsf{W}\}$$, a tilt is defined as the process whereby each polyomino is translated maximally in $d$: it slides iteratively, moving one unit per step if that move does not lead to wall-collision or overlap, until blocked. All polyominoes act in parallel, and the process iterates until the configuration is $d$-terminal (no piece can move further in $d$). This "maximal move per command" mechanism—contrasting sharply with traditional, local, single-agent models—is the distinguishing feature of the full-tilt regime (Avila-Jimenez et al., 6 Dec 2025, Blumenberg et al., 2023, Balanza-Martinez et al., 2019).

If the system supports glues or edge labels, adjacent polyominoes may "stick" and merge if their facing glue types (drawn from a finite alphabet $\Sigma$) match according to a symmetric glue function $G: \Sigma \times \Sigma \rightarrow \{0,1\}$; this induces permanent bonding, yielding larger moving assemblies ("polyominoes") with subsequent tilts.

2. Universal Assembly and Construction Results

The full-tilt model enables both shape-specific and universal self-assembly of target patterns and classes of polyominoes.

Strong and Weak Universality

• Strong Universality: There exist initial configurations of $O(hw)$ tiles and $O(hw)$ walls such that any binary-labeled $h\times w$ pattern can be realized by a finite sequence of tilts, with no extra particles remaining (strong representation) (Balanza-Martinez et al., 2019). The construction uses multiple chambers (fuel, loading, construction), and carefully cycles tiles to the correct position row by row, with each global tilt driving a necessary transport or assembly action.
• Weak Universality: Relaxing the requirement of exact coverage (allowing unbonded helper tiles), one can build every connected shape within an $h\times w$ box using $O(hw)$ tiles and tilts via similar multi-chamber machinery, filling only target locations with sticky tiles and leaving "sand" elsewhere (Balanza-Martinez et al., 2019).
• Drop Shapes: For the "drop shape" class—shapes that can be assembled by a seed and a sequence of single-tile "drops" from any of the four cardinal directions—a construction of size $O(h^2w)$ and $O(h^2w)$ tilts suffices for strong universality (Balanza-Martinez et al., 2019).

These universal constructions exploit the regularity and determinism of full-tilt actuation to realize large classes of polyomino assemblies efficiently, bridging programmable matter, tiling theory, and collective robotics (Balanza-Martinez et al., 2019, Blumenberg et al., 2023).

3. Computational Complexity of Motion Planning

Central reachability problems—relocation, occupancy, vacancy, and reconfiguration—are PSPACE-complete even under powerful deterministic actuation such as periodic rotation cycles (Avila-Jimenez et al., 6 Dec 2025, Balanza-Martinez et al., 2019, Blumenberg et al., 2023).

• Occupancy: Decide if some cell can eventually be occupied by any tile under some sequence of tilts.
• Relocation: Decide if a specified polyomino can be moved to a designated target location via tilts.
• Reconfiguration: Decide if a given configuration $C$ can be transformed into another configuration $C'$.
• Vacancy: Decide if a cell can be emptied.

PSPACE-hardness is typically proved by reduction from space-bounded Turing machine computation, using tile and board gadgets to simulate transitions and encode tape, state, and head. Notably, such results hold even with just singleton tiles plus a single domino and in the presence of deterministic, periodic tilt sequences (e.g., a fixed 4-cycle $$\langle \mathsf{N},\mathsf{E},\mathsf{S},\mathsf{W}\rangle$$), leveraging the capacity of the interaction geometry to encode computational flow (Avila-Jimenez et al., 6 Dec 2025). All such problems are in PSPACE since configurations can be represented in polynomial space, and the configuration graph can be explored non-deterministically (Balanza-Martinez et al., 2019, Blumenberg et al., 2023, Avila-Jimenez et al., 6 Dec 2025).

4. Algorithmic and Planning Approaches

Algorithmic research in the full-tilt model has focused on both optimal and scalable heuristics for polyomino assembly, shape construction, and general reachability.

• Best-First with Global Heuristics: The configuration graph is searched using A* or greedy best-first planners with admissible heuristics such as $h_{GD}(C,X)$—the minimum grid distance of the $n^{th}$ reachable tile to $X$—with pruning rules based on reachability and packing constraints (Blumenberg et al., 2023).
• Incremental Single-Tile Strategies: Inspired by assembly theory, tiles are brought one-by-one into exposed positions adjacent to a target polyomino. This uses cost-bounding heuristics such as $h_{MMP}$ and $h_{DFP}$ (distance-to-frontier estimates) to guide search toward partial assemblies (Blumenberg et al., 2023).
• Sampling-based Planners: Algorithms akin to rapidly-exploring random trees (RRT) randomly sample configurations, attempting to connect via short tilt paths, often with goal biasing and distance metrics adapted to global-tilt transitions (Blumenberg et al., 2023).
• Gadget-based Universal Simulation: For general computation, placement and shape of tile gadgets encode the state-space and transition function of a Turing machine, driven by deterministic tilt sequences so that computation is carried out spatially (Avila-Jimenez et al., 6 Dec 2025).

Empirical studies show the trade-offs between solution optimality, computational runtime, and scalability, with incremental and greedy heuristics outperforming naive simultaneous strategies on larger instances (Blumenberg et al., 2023).

5. Simulation of General Computation and Circuit Models

A key discovery is that the full-tilt model admits direct, efficient simulation of universal Turing machines and threshold circuits via geometric construction:

• One-Tape TM Simulation: For an $s$-state Turing machine of tape-length $n$, $O(ns^3)$ tiles and matching board suffice, and each computational step can be simulated by $O(1)$ repetitions of a fixed 4-tilt cycle. Gadget "cells" encode tape position, read/write logic, state decode/update, and left/right movement, with restricted tile types (singletons and a single domino) (Avila-Jimenez et al., 6 Dec 2025).
• Threshold Circuits: Unbounded-fan-in Majority gates, AND, and OR gates can be simulated in $O(1)$ tilts per logic level; a circuit of depth $d$ can be emulated in $O(d)$ rotations and $O(w^2d)$ board area for width $w$ (Avila-Jimenez et al., 6 Dec 2025).
• Extensions: Asymptotic simulation complexity applies to two-tape TMs ($O(1)$ tilts per step), systolic array models, and programming of tape state by preparatory tilt sequences.

By implementing both the logic and control signals geometrically, full-tilt motion planning subsumes large classes of symbolic computation and digital hardware models.

6. Experimental Evaluation and Algorithmic Benchmarks

Extensive evaluations explore algorithmic performance on procedurally generated instances, systematically varying board size, target shape, excess tile count, glue complexity, and presence/absence of a fixed seed. Key metrics are:

• Success Rate: Fraction of instances solved within time limits.
• Solution Length: Number of tilt operations in a successful plan.
• Runtime and Memory: Wall-clock time and peak resource use.
• Search Nodes: Total generated configurations (Blumenberg et al., 2023).

Salient findings:

• Greedy heuristics deliver fastest runtimes with longer, suboptimal solutions.
• Incremental search with targeted heuristics scales best for large $n$ and complex boards.
• Fixed-seed (immobile anchor) variants are more efficiently solvable, e.g., $90\%$ of seeded instances resolved in $\ll$1 s with DFP heuristics.
• Sampling-based planners (RRT) are memory-efficient but sensitive to instance size.
• Only greedy, incremental, and fixed-seed-specific heuristics are practical for large-scale full-tilt planning (Blumenberg et al., 2023).

7. Significance and Theoretical Implications

The full-tilt model unites algorithmic motion planning, formal self-assembly, programmable matter, and computational universality via deterministic collective actuation. Foundational results include:

• PSPACE-completeness of all major reachability and assembly variants, even under deterministic, periodic tilt control and minimal particle types (Avila-Jimenez et al., 6 Dec 2025, Balanza-Martinez et al., 2019, Blumenberg et al., 2023).
• Existence of universal constructors for general patterns and shapes with polynomial-time assembly under global tilts (Balanza-Martinez et al., 2019).
• Feasibility of efficient simulation of arbitrary space-bounded Turing machines and threshold circuits, making the model a substrate for programmable computation using mechanical means (Avila-Jimenez et al., 6 Dec 2025).
• Algorithmic advances in heuristics, configuration graph search, and sampling, supporting practical planning under severe actuation constraints (Blumenberg et al., 2023).

The full-tilt abstraction forms a rigorous touchstone for research in globally actuated robot swarms, algorithmic self-assembly, complexity in motion planning, and universal geometry-driven computation.