Thicket data repository for the EEG
anil.recoil.org
1{
2 "id": "https://gabrielmahler.org/walkability/ai/ml/2025/06/07/evaluation3",
3 "title": "Walkability Chapter 5: Evaluation (Part 3: Path Generation)",
4 "link": "https://gabrielmahler.org/walkability/ai/ml/2025/06/07/evaluation3.html",
5 "updated": "2025-06-07T10:40:11",
6 "published": "2025-06-07T10:40:11",
7 "summary": "Semantically-based Path Generation",
8 "content": "<h1>Semantically-based Path Generation</h1>\n\n<p>With the semantically-based point-wise scores for general walkability from the last part\nand the four specific objectives (greenery, shopping, historicity, and\nsafety), we revisit the exemplary routing scenarios. We\ndemonstrate that our pipeline not only has the ability to generate more\nwalkable paths but, unlike the baseline frameworks, can also leverage\npeculiar routing opportunities. We show how our approach can be used to\naddress the shortcomings of existing frameworks, and\n<strong>improve the quality of path-finding solutions</strong>, answering our second\nresearch question.</p>\n\n<h3>Pedestrian-friendly Alternatives in the Supermarket-adjacent Walk</h3>\n\n<p>In the first routing problem (defined by the park-adjacent residential\nareas), it is primarily the general walkability and greenery objectives\nthat resonate with the presented context.</p>\n\n<p>Both greenery and general\nwalkability paths use highly walkable park segments adjacent to a river.\nThis is in stark contrast to the open-source baselines, where the generated paths\ntangle through homogeneous residential blocks (OSRM, GraphHopper), or\nmerely follow busy but straightforward roads (Valhalla). A similar route\nto Valhalla’s (although guided by a very different rationale) is\ngenerated under the shopping objective and pursues proximity to stores\nthat generally lie on the same streets Valhalla targets.</p>\n\n<p><img alt=\"alt text\" src=\"https://gabrielmahler.org/assets/images/thesis/new%20images/parkside%20lidl/cases%20parkside%20lidl.jpeg\"></p>\n\n<p>Moreover, the case for our preference-based paths is further supported\nby the total lengths of the generated paths. The considerably more pleasant\nwalkability and greenery-focused paths remain only slightly longer than\nthe OSRM or GraphHopper alternatives. Furthermore, the shopping, safety,\nand historically-focused paths are even shorter.</p>\n\n<p><strong>Table: Total distance of each path generated for the “supermarket-adjacent walk”</strong></p>\n\n\n\n \n \n Algorithm\n Distance (meters)\n \n \n \n \n <em>GraphHopper</em>\n 1974.06\n \n \n <em>OSRM</em>\n 2008.08\n \n \n <em>Valhalla</em>\n 1805.71\n \n \n <em>General Walkability</em>\n 2241.03\n \n \n <em>Greenery</em>\n 2164.63\n \n \n <em>Historical</em>\n 1884.81\n \n \n <em>Safety</em>\n 1877.08\n \n \n <em>Shopping</em>\n 1801.81\n \n \n\n\n<h3>Maximizing Green Paths in the Long City-spanning Walk</h3>\n\n<p>Next, we revisit the spatially significant routing scenario. As discussed, here (similarly\nto the first scenario), path-finding is provided with an opportunity to\nencapsulate much of the prolonged tour into highly walkable\nenvironments, particularly parks. Nonetheless, as discussed earlier, the open-source frameworks\nlargely ignore this opportunity and generate a path leading through\nmostly residential areas of underwhelming walkability.</p>\n\n<p><img alt=\"alt text\" src=\"https://gabrielmahler.org/assets/images/thesis/new%20images/random%20long/cases%20random%20long.jpeg\"></p>\n\n<p>This, however, is not the case for our path generated under the greenery\nobjective. This path maximizes the duration of\ngreen spaces, only diverging when necessary. In contrast, as the general\nwalkability scores are defined by a broader spectrum of factors, the\npath generated under their umbrella combines walkable urban areas with\ngreen spaces. The remaining three objectives (shopping, historicity, and\nsafety) attempt to leverage the sparse relevant elements present in the\nadjacent residential areas. However, reflecting the character of the\nenvironment, none of the outputs present discussion-worthy features.</p>\n\n<p><strong>Table: Total distance of each path generated for the “long city-spanning walk”</strong></p>\n\n\n\n \n \n Algorithm\n Distance (meters)\n \n \n \n \n <em>GraphHopper</em>\n 8284.38\n \n \n <em>OSRM</em>\n 8165.23\n \n \n <em>Valhalla</em>\n 8257.62\n \n \n <em>General Walkability</em>\n 8568.04\n \n \n <em>Greenery</em>\n 9103.54\n \n \n <em>Historical</em>\n 8838.61\n \n \n <em>Safety</em>\n 7928.67\n \n \n <em>Shopping</em>\n 7928.67\n \n \n\n\n<p>The (atypically for urban pedestrian path-finding) large problem space\nis also reflected in the margins of the generated paths’ overall lengths. In the case of the greenery,\nthe resulting path exceeds the open-source baselines by almost a\nkilometer. Nonetheless, considering this specific scenario, we argue\nthat such a difference is still within reason.</p>\n\n<h3>Urban-Centered Variants in the Greenbelt Walk</h3>\n\n<p>In the next scenario, we study how our framework manages to prioritize\ndiverse urban options over more greenery-based\nalternatives. As discussed earlier, using the open-source frameworks\nhere overwhelmingly resulted in paths prioritizing the latter.</p>\n\n<p><img alt=\"alt text\" src=\"https://gabrielmahler.org/assets/images/thesis/new%20images/ph%20cst%20short/ph%20cst%20cases.jpeg\"></p>\n\n<p>Ultimately, so did the paths generated under our walkability and\ngreenery objectives. Our historical, safety, and\nshopping-related paths, nevertheless, produced much more interesting\nsolutions. The shopping and safety-focused path maximizes the duration\nin the city, leveraging busy pedestrian segments surrounded by stores\nand other establishments. The historical path follows the same\ntrajectory, before making a sharp turn to leverage a path through a\nlate-Gothic university college.</p>\n\n<p>Furthermore, as highlighted, the distances of the\npreference-based paths once again remain within a reasonable margin from\nthe duration-optimizing baselines.</p>\n\n<p><strong>Table: Total distance of each path generated for the “greenbelt walk”</strong></p>\n\n\n\n \n \n Algorithm\n Distance (meters)\n \n \n \n \n <em>GraphHopper</em>\n 2160.66\n \n \n <em>OSRM</em>\n 2095.53\n \n \n <em>Valhalla</em>\n 2088.55\n \n \n <em>General Walkability</em>\n 2307.68\n \n \n <em>Greenery</em>\n 2304.71\n \n \n <em>Historical</em>\n 2080.30\n \n \n <em>Safety</em>\n 2144.18\n \n \n <em>Shopping</em>\n 2143.16\n \n \n\n\n<h3>Opportunities in the City Center Walk</h3>\n\n<p>In the fourth scenario, characterized by a dense urban environment and a\nmore modest spatial dimension, we demonstrate more nuanced abilities of\nour approach, particularly through the shopping-focused objective. While\n(particularly in this scenario) the paths generated by the open-source\nbaselines served merely efficiency\npurposes, our shopping-focused assessment identified a nearby shopping\nmall as an area of strong interest, and subsequently situated its path\nto lead through it. Additionally, the walkability and\ngreenery-focused objectives took advantage of a short segment adjacent\nto a park, and the safety-optimizing path remained close to busy roads,\navoiding quieter roads and alleys.</p>\n\n<p><img alt=\"alt text\" src=\"https://gabrielmahler.org/assets/images/thesis/new%20images/ps%20senate/ph%20senate%20cases.jpeg\"></p>\n\n<p>Our preference-based paths remained close to the open-source baselines\nin terms of overall distance. In fact, in this scenario, the\nlongest road was generated by the Valhalla framework.</p>\n\n<p><strong>Table: Total distance of each path generated for the “city center walk”</strong></p>\n\n\n\n \n \n Algorithm\n Distance (meters)\n \n \n \n \n <em>GraphHopper</em>\n 921.93\n \n \n <em>OSRM</em>\n 1099.31\n \n \n <em>Valhalla</em>\n 1106.36\n \n \n <em>General Walkability</em>\n 946.31\n \n \n <em>Greenery</em>\n 904.73\n \n \n <em>Historical</em>\n 899.96\n \n \n <em>Safety</em>\n 920.22\n \n \n <em>Shopping</em>\n 940.45\n \n \n\n\n<h3>Safer-focused Options in the Suburban Stretch Walk</h3>\n\n<p>Finally, we return to our last scenario situated within a dense but\nrather homogeneous residential area. Reflecting on the results generated\nby the open-source frameworks (particularly GraphHopper and OSRM), we\nuse the safety-based objective to generate a contrasting path, mostly\nfollowing a busy (and, by assumption, safer) road. As a segment, the most associated\nwith relevant geospatial nodes, this road is also prioritized by the\nshopping-focused path. It must be noted that, once again, our shopping\nand safety-related paths share many similarities with the route\ngenerated by Valhalla. However, once again, we attribute this primarily\nto Valhalla’s attention to strictly segment-related features, such as\njunctions and crossings, that are entirely ignored in our method.\nSimilarly, our general walkability, greenery, and historically-based\npreferences are projected in slightly longer paths, due to their various attempts\nto maximize their objective in a highly homogeneous environment.</p>\n\n<p><img alt=\"alt text\" src=\"https://gabrielmahler.org/assets/images/thesis/new%20images/residential/residential%20cases.jpeg\"></p>\n\n<p><strong>Table: Total distance of each path generated for the “suburban stretch walk”</strong></p>\n\n\n\n \n \n Algorithm\n Distance (meters)\n \n \n \n \n <em>GraphHopper</em>\n 2140.88\n \n \n <em>OSRM</em>\n 2173.51\n \n \n <em>Valhalla</em>\n 2071.77\n \n \n <em>General Walkability</em>\n 2523.46\n \n \n <em>Greenery</em>\n 2539.48\n \n \n <em>Historical</em>\n 2265.97\n \n \n <em>Safety</em>\n 2146.01\n \n \n <em>Shopping</em>\n 2146.01\n \n \n\n\n<h2>Discussion</h2>\n\n<p>In this chapter, we answer and use our pipeline to provide solutions to\nour three initial questions. We present specific routing examples\nthat showcase common shortcomings of popular path-finding frameworks,\nparticularly as they disregard highly walkable routing opportunities. We\nleverage our point-wise assessment and path-finding pipeline to generate\nwalkability and preference-based routes, addressing the stated\nshortcomings and improving upon the existing methods. We show that our\npipeline can accurately identify urban points of particular interest,\nand our path-finding search can effectively utilize them in its\nsolutions. Finally, we highlight the accessibility to user-specific\npreferences, enabled with our natural language-based approach. In sharp\ncontrast to existing frameworks, preferences in our framework are\ndefined with plain sentences.</p>\n\n<p>Besides illustrating a novel method to pedestrian path-finding, our\nexamples also highlight the importance of public geospatial datasets,\nparticularly OSM. As demonstrated, the quality of our framework’s\noutputs is extremely dependent on the accuracy and descriptiveness of\nthe data provided by OSM. This is particularly well illustrated by our\nsafety-focused objective, which relies on a very specific selection of\ngeospatial elements. Nonetheless, besides the many benefits, the\nopen-source nature of OSM also imposes its sets of various dangers, and\nactively contributing communities are essential to prevent dishonest use\nof these resources. Particularly in frameworks such as ours, unbalanced\ndata coverage or intentionally misleading records can lead to skewed\noutputs, potentially helping malicious (e.g., profit-driven) intents.\nNonetheless, such dangers can be minimized by active community-led\nrevisions of the datasets.</p>",
9 "content_type": "html",
10 "author": {
11 "name": "",
12 "email": null,
13 "uri": null
14 },
15 "categories": [
16 "Walkability",
17 "AI/ML"
18 ],
19 "source": "https://gabrielmahler.org/feed.xml"
20}