kth.sePublications KTH
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Parameter-Robust MPPI for Safe Online Learning of Unknown Parameters
KTH, School of Electrical Engineering and Computer Science (EECS), Centres, Digital futures. KTH, School of Electrical Engineering and Computer Science (EECS), Intelligent systems, Robotics, Perception and Learning, RPL.ORCID iD: 0000-0001-6046-7460
Massachusetts Institute of Technology, Reliable Autonomous Systems Lab, Cambridge, MA, USA.ORCID iD: 0000-0003-4177-3010
Swiss Federal Institute of Technology in Zürich, Automatic Control Laboratory, Switzerland.
KTH, School of Electrical Engineering and Computer Science (EECS), Intelligent systems, Robotics, Perception and Learning, RPL. KTH, School of Electrical Engineering and Computer Science (EECS), Centres, Digital futures.ORCID iD: 0000-0003-4173-2593
Show others and affiliations
2026 (English)In: IEEE Robotics and Automation Letters, E-ISSN 2377-3766, Vol. 11, no 4, p. 3931-3938Article in journal (Refereed) Published
Abstract [en]

Robots deployed in dynamic environments must remain safe even when key physical parameters are uncertain or change over time. We propose Parameter-Robust Model Predictive Path Integral (PRMPPI) control, a framework that integrates online parameter learning with probabilistic safety constraints. PRMPPI maintains a particle-based belief over parameters via Stein Variational Gradient Descent, evaluates safety constraints using Conformal Prediction, and optimizes both a nominal performance-driven and a safety-focused backup trajectory in parallel. This yields a controller that is cautious at first, improves performance as parameters are learned, and ensures safety throughout. Simulation and hardware experiments demonstrate higher success rates, lower tracking error, and more accurate parameter estimates than baselines.

Place, publisher, year, edition, pages
Institute of Electrical and Electronics Engineers (IEEE) , 2026. Vol. 11, no 4, p. 3931-3938
Keywords [en]
Model Learning for Control, Robot Safety
National Category
Robotics and automation Computer Sciences Control Engineering
Identifiers
URN: urn:nbn:se:kth:diva-377643DOI: 10.1109/LRA.2026.3662531ISI: 001696543000008Scopus ID: 2-s2.0-105029919703OAI: oai:DiVA.org:kth-377643DiVA, id: diva2:2042826
Note

QC 20260303

Available from: 2026-03-03 Created: 2026-03-03 Last updated: 2026-05-29Bibliographically approved
In thesis
1. Risk-aware Robot Safety via Control in Belief Space and Beyond
Open this publication in new window or tab >>Risk-aware Robot Safety via Control in Belief Space and Beyond
2026 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

Robotic systems must operate safely despite noisy measurements, partial observability, and imperfect models of their dynamics. These sources of uncertainty fundamentally challenge how safety can be ensured, as classical control methods typically assume exact knowledge of the system state and model. This thesis develops a principled foundation for robot safety under uncertainty by designing control strategies directly in belief space, a representation that captures how uncertainty evolves through stochastic motion and observation processes. Working in belief space enables safety and performance requirements to be expressed in terms of the robot's probabilistic description of the state, rather than an assumed deterministic one.

Viewing autonomy through this lens enables explicit reasoning about risk and information. Safety specifications can be expressed as risk constraints on the belief, allowing the controller to account for low-probability but safety-critical tail events. At the same time, the belief representation enables the robot to reason about how observations can reduce uncertainty, and to actively steer toward regions where uncertainty can be reduced more effectively. A key contribution of this thesis is the formalization of control certificates such as Control Barrier Functions and Control Lyapunov Functions in belief spaces. These certificates provide formal safety and convergence guarantees directly in belief space while admitting computationally tractable controllers.

The thesis further extends these insights beyond belief space control. It interprets components of robot control such as trajectory planning and certificate generation as dynamical processes whose evolution can themselves be subject to invariance principles. This broader viewpoint leads to new formulations that treat trajectory generation and safety verification within a unified dynamical-systems framework.

Together, these contributions advance the ability of autonomous systems to reason about and act safely under uncertainty, supporting reliable deployment in real-world environments.

Abstract [sv]

Robotsystem måste fungera säkert trots brusiga mätningar, partiell observerbarhet och ofullständiga modeller av sin dynamik. Dessa osäkerhetskällor utmanar hur säkerhet kan garanteras, eftersom klassiska styrmetoder vanligtvis antar exakt kunskap om systemets tillstånd och modell. Denna avhandling utvecklar en principiell grund för robotsäkerhet under osäkerhet genom att utforma styrstrategier direkt i rymden av tillståndsfördelningar, en representation som fångar hur osäkerhet utvecklas genom stokastiska rörelse- och observationsprocesser. Att arbeta i denna rymd gör det möjligt att formulera säkerhets- och prestandakrav i termer av robotens probabilistiska beskrivning av tillståndet, snarare än ett deterministiskt sådant.

Detta perspektiv möjliggör ett explicit resonemang kring risk och information. Säkerhetsspecifikationer kan uttryckas som riskbegränsningar på tillståndsfördelningar, vilket gör att regulatorn kan ta hänsyn till osannolika men säkerhetskritiska händelser. Samtidigt gör representationen det möjligt för roboten att resonera kring hur observationer kan minska osäkerheten och aktivt styra mot områden där den kan lokalisera sig bättre. Ett centralt bidrag i denna avhandling är formaliseringen av kontrollcertifikat såsom Control Lyapunov och Barrier Functions i fördelningsrymden. Dessa certifikat ger formella garantier för säkerhet och konvergens direkt i denna rymd, samtidigt som de möjliggör beräkningsmässigt hanterbara regulatorer.

Avhandlingen utvidgar dessutom dessa insikter bortom reglering baserad på tillståndsfördelningar. Komponenter i robotstyrning, såsom trajektorieplanering och generering av certifikat, tolkas som dynamiska processer vars utveckling kan omfattas av invariansprinciper. Detta bredare perspektiv leder till formuleringar som behandlar trajektoriegenerering och säkerhetsverifiering inom ett enhetligt ramverk av dynamiska system.

Tillsammans bidrar dessa resultat till att förbättra autonoma systems förmåga att resonera och agera säkert under osäkerhet, och stödjer därmed en tillförlitlig användning i verkliga miljöer.

Place, publisher, year, edition, pages
KTH Royal Institute of Technology, 2026. p. 79
Series
TRITA-EECS-AVL ; 2026:49
National Category
Robotics and automation
Research subject
Computer Science
Identifiers
urn:nbn:se:kth:diva-381048 (URN)978-91-8106-614-2 (ISBN)
Public defence
2026-06-05, D3, Lindstedtsvägen 5, plan 3, KTH Campus, Stockholm, 14:00 (English)
Opponent
Supervisors
Note

QC 20260508

Available from: 2026-05-08 Created: 2026-05-08 Last updated: 2026-05-19Bibliographically approved

Open Access in DiVA

No full text in DiVA

Other links

Publisher's full textScopus

Authority records

Vahs, MattiTumova, Jana

Search in DiVA

By author/editor
Vahs, MattiChoi, JaeyounTumova, JanaFan, Chuchu
By organisation
Digital futuresRobotics, Perception and Learning, RPL
In the same journal
IEEE Robotics and Automation Letters
Robotics and automationComputer SciencesControl Engineering

Search outside of DiVA

GoogleGoogle Scholar

doi
urn-nbn

Altmetric score

doi
urn-nbn
Total: 44 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf