TSP Solver – User Guide

Move panels

• Canvas: drag by the border/edge (cursor changes to “move”). Double‑click the edge to reset position.
• Route Details: drag using the grip bar at the top (appears on hover).

Resize panels

• Use the browser resize handle (bottom‑right) on both Canvas and Route Details.
• For huge datasets, enable Skip Drawing in Visual Config to keep the UI responsive.

Method selection

• Very large point sets: Robopol Fast (Turbo / Superboost).
• Large (1,000–10,000): Robopol Fast or LKH (time/quality trade‑off).
• Medium (100–1,000): Robopol Precise or LKH.
• Small/medium with a quality-focused goal: LKH + Robopol Refined.
• Fast high-quality hybrid: LKH + Robopol Escape Polish with reduced Refined-style parameters.

Robopol Refined – tuned parameters (for n < 1000)

• Num Runs > 1
• Beam Width > 3
• ILS Iterations ≥ 2000
• Perturbation Candidates = 16
• Candidate Edges/Node = 10

For the LKH + Robopol hybrid methods, these values can often be reduced because LKH already provides a high-quality starting route. This can keep quality high while shortening calculation time.

Choose Enable Edge-Guided Search based on your goal: enabled = much faster, but potentially less precise; disabled = slower heavy-style search, but often better when you need maximum route quality.

New Project

Resets to the initial state (clears canvas/output, keeps settings).

Save As

Saves the project as .json (air: tsp_air_data.json, road: tsp_road_data.json).

Open Project

Loads a saved .json project created by TSP Solver.

Import / Export Image

Import a background template and export the current visualization as an image.

Export SVG

Exports route + points as scalable vector graphics (.svg).

Visual config

• Canvas colors: Background, Node, Link
• Sizes: Node size, Link width
• Skip Drawing (large datasets): disables rendering of points and route on the canvas
• Page colors (UI theme): Page BG, Sidebar, Panels, Accent, Accent Dark, Text, Borders, Inputs BG, etc.
• Reset Page Colors: restores default TSP Solver theme colors

Method config

Use Method Config to tune algorithm parameters. For Robopol Refined, start from the recommended values in the “Settings (recommended)” section.

Robopol Refined parameters (reference)

The items below match the Method Config inputs used by Robopol Refined. Descriptions are based on the actual implementation (see app_2025.py).

• ILS Iterations (ils_iterations): number of main ILS loop iterations. Higher = more time, usually better quality (diminishing returns).
• No Improvement Limit (ils_no_improvement_limit): stop after this many ILS iterations without improving the global best. Also influences internal “stagnation” thresholds.
• Number of Runs (refined_runs): independent restarts; the best run is returned. More runs = higher chance of improvement, higher CPU usage (runs are parallelized).
• Enable Edge-Guided Search (refined_edge_guided_enabled): controls the main speed/precision trade-off in Refined. Enabled prioritizes promising tour/candidate/hot/reduction edges and suppresses many broad fallback perturbations, so it is much faster but can be less precise. Disabled behaves closer to the older heavy mode, searches a larger space, and can produce a better route when quality matters more than runtime.
• Beam Width (beam_width): how many candidate solutions are kept in the beam at once. Higher = more exploration and CPU cost.
• Perturbation Candidates (num_candidates): how many perturbed candidates are generated per ILS iteration before selecting the best ones for the beam.
• Targeting Probability (targeting_probability): probability (0–1) of using a targeted perturbation strategy vs. a more random one.
• Candidate Edges/Node (num_candidate_edges): candidate-set width for local search (how many “promising neighbors” each node considers). Higher can improve quality but increases preprocessing and local search cost.
• Max Segment Pct (max_segment_pct): maximum segment length for segment-move perturbations as a fraction of \(n\) (e.g. 0.2 = up to ~20% of nodes).
• Tolerance Base (acceptance_tolerance_base): controls acceptance of non-improving moves (exploration). Higher = more “willing to accept worse” (can help escape local minima but can also slow convergence).
• ILS 2-opt Iterations (ils_2opt_iterations): iteration budget for the 2-opt phase inside ILS local search steps.
• ILS Move Iterations (ils_move_iterations): iteration budget for the segment-move local search inside ILS (used for “polish” moves).
• ILS Max Segment Size (ils_max_segment_size): maximum moved segment length (in nodes) for segment-move local search.
• ILS Max Nearest Neighbors (ils_max_nearest_neighbors): how many nearest neighbors are considered during local search operations (speed/quality trade-off).
• ILS Deep Search (ils_deep_search): deeper search setting for segment-move operations (more thorough = slower).
• Max Local Search Attempts (max_local_search_attempts): cap on repeated local search attempts per ILS step.
• Local Search Threshold (local_search_no_improvement_threshold): stop local search early after this many consecutive attempts without improvement.

LKH Multi-vehicle (mTSP) settings

TSP Solver supports multi-vehicle routing (mTSP) via LKH-3.0.10 from the Method Config panel. It is intended mainly for AIR 2D and ROAD workflows. Use AIR 3D (x,y,z) carefully and test smaller jobs before large runs.

• Problem type: choose TSP (1 route) or Multi-vehicle (mTSP).
• Vehicles (vehicles): number of vehicles/salesmen (routes) to split the tour into.
• Objective (mtsp_objective):
  • MINSUM: minimize total sum of all routes (best overall distance/time).
  • MINMAX: minimize the longest single route (fairness by distance/time).
  • MINMAX_SIZE: balance workload by minimizing the maximum number of stops per vehicle (not a direct distance objective).
• Depot index (1..n) (depot_index): start/end node for all routes (1 = first point).
  ROAD note: in Road mode this index refers to the order of successfully geocoded addresses (1 = first). Failed address lines are skipped.
• Min points / vehicle (mtsp_min_size): minimum number of stops per vehicle (excludes depot). Use to prevent 1-vs-99 splits.
• Max points / vehicle (mtsp_max_size): maximum number of stops per vehicle (excludes depot). Use to cap workload.
• Road optimize by (road_objective): choose Distance (km) or Time (sec) for ROAD mode.

LKH-3.0.10 advanced settings

These controls are under LKH-3.0.10 METHOD → ADVANCED SETTINGS in Method Config. They tune the LKH engine used by plain LKH and by LKH-based workflows where applicable.

• MOVE_TYPE (moveType, default 5): maximum Lin-Kernighan move depth. Higher values search more neighborhoods and can increase runtime.
• PATCHING_C (patchingC, default 0): cycle patching setting after LK moves. Higher values can make the search more thorough but slower.
• PATCHING_A (patchingA, default 0): patching breadth/alternate patching setting. Keep at 0 unless you are testing LKH quality tuning.
• RUNS (runs, default 1): independent LKH runs; the best result is used. More runs usually means more time.
• MAX_TRIALS (maxTrials, default 1): trial budget inside each run. Increase for a more thorough LKH search.
• CANDIDATE_SET_TYPE (candidateSetType, default DELAUNAY): candidate-edge generator. UI options are DELAUNAY, ALPHA, DELAUNAY PURE, NEAREST-NEIGHBOR, QUADRANT, and POPMUSIC.
• MAX_CANDIDATES (maxCandidates, default 5): maximum candidate edges per node. Higher values can improve search coverage but increase memory/runtime.
• INITIAL_PERIOD (initialPeriod, default 100): LKH preprocessing/ascent period used while building candidate information.
• POPMUSIC_SOLUTIONS (popmusicSolutions, default 50): number of POPMUSIC candidate solutions when POPMUSIC candidate sets are used.

Add points

Click on the canvas to add points. Point #1 is used as the start/end.

Delete points

Enable delete mode and click a point to remove it.

Remove lines

Clears the currently drawn route connections.

Clear all

Removes all points, routes and imported background images.

Calculate

Computes the route for the selected method. Requires at least 2 points.