MétaCan
Menu
← Back to cohort
Record W7043091407

Reinforcement Learning for Stair Climbing with Ascento: A Two-Wheeled Legged Robot

2024· other· fr· W7043091407 on OpenAlexfundno aff

Bibliographic record

VenuePolyPublie (École Polytechnique de Montréal) · 2024
Typeother
Languagefr
Field
Topic
Canadian institutionsnot available
FundersPolytechnique Montréal
KeywordsWizard of ozLegged robotPoison control
DOInot available

Abstract

fetched live from OpenAlex

RÉSUMÉ: Ascento Robotics, une entreprise de sécurité, utilise un robot à roues et à pattes pour effectuer des patrouilles de sécurité dans de vastes espaces extérieurs. Actuellement, Ascento est capable de naviguer sur un terrain plat et est robuste face à de légers obstacles. Cependant, une limitation critique du système actuel est l’incapacité du robot à monter des escaliers, ce qui limite sa fonctionnalité dans les environnements à plusieurs étages. Ce mémoire propose une nouvelle méthode pour surmonter cette limitation, en utilisant l’Apprentissage par Renforcement (RL) pour apprendre un contrôleur en simulation. L’approche proposée adopte une formulation de tâche RL basée sur la position, en contraste avec les contrôleurs existants basés sur la vitesse. Elle exploite la structure asymétrique acteur-critique pour utiliser des informations privilégiées provenant du simulateur pendant l’entraînement, tout en éliminant le besoin d’avoir accès à ces données pendant le déploiement. Cette approche simplifie considérablement les exigences opérationnelles du robot, car elle ne repose sur aucune information de capteur extéroceptif. Une nouvelle observation booléenne introduite dans le contrôleur sert de mode d’escalade d’escalier pouvant être activé ou désactivé. Cette recherche présente également les résultats du processus de transfert de la simulation au monde réel, en utilisant des techniques telles que la randomisation de domaine, ainsi qu’une généralisation de la méthode présentée et son application à d’autres robots. Le contrôleur résultant, qui ne nécessite pas d’informations de capteur ni de système de positionnement externe, permet au robot de monter avec succès des marches de 15 cm, une tâche qui était auparavant impossible pour le robot Ascento. Cette avancée élargit le champs opérationnel du robot, améliorant son potentiel dans les applications de sécurité. ABSTRACT: Ascento Robotics, a security company, utilizes a wheeled-legged robot to do security patrols in large outdoor areas. Ascento currently has the capacity to navigate flat terrain and is robust to minor obstacles. However, one critical limitation of the current system is the robot’s inability to climb stairs, limiting its functionality in multi-story settings. This thesis proposes a novel method to overcome this limitation, employing Reinforcement Learning (RL) to train a controller in simulation. The proposed approach adopts a position-based RL task formulation, contrasting with the existing velocity-based control. It leverages the asymmetric actor-critic structure to utilize privileged information from the simulator during training while eliminating the need for this data during deployment. This approach significantly simplifies the operational requirements of the robot, since it does not rely on any exteroceptive sensor information. A new boolean observation introduced to the controller serves as a stairclimbing mode that can be activated or deactivated. This research also presents findings from the sim-to-real transfer process, using techniques such as domain randomization. The resultant controller, which does not require sensor information or a positioning system, successfully enables the robot to climb 15 cm steps, a task that was previously impossible for Ascento. This advancement broadens the operational terrain for the robot, enhancing its potential in security applications.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.

How this classification was reachedexpand

Full frame machine prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.

metaresearch head score (Codex)0.000
metaresearch head score (Gemma)0.001
Version: metacan-v3-hybrid-931329e0061cValidation status: machine_predicted_unvalidated
Candidate categoriesnone
Consensus categoriesnone
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Simulation or modeling · Consensus signal: Simulation or modeling
GenreCandidate signal: Empirical · Consensus signal: none
Teacher disagreement score0.006
Threshold uncertainty score0.013

Distilled classifier scores by category (both heads)

CategoryCodexGemma
Metaresearch0.0000.001
Meta-epidemiology (narrow)0.0010.000
Meta-epidemiology (broad)0.0000.000
Bibliometrics0.0000.000
Science and technology studies0.0000.001
Scholarly communication0.0000.000
Open science0.0010.001
Research integrity0.0010.001
Insufficient payload (model declined to judge)0.0030.000

Machine scores (provisional)

The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.

Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.

Opus teacher head0.012
GPT teacher head0.254
Teacher spread0.242 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it

Classification

machine, unvalidated

Machine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.

The models applied no category: nothing in the taxonomy fit this work.
Study designSimulation or modeling
Domainnot available
GenreEmpirical

How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".

Quick stats

Citations0
Published2024
Admission routes1
Has abstractyes

Explore more

Same venuePolyPublie (École Polytechnique de Montréal)→French-language works237,207→