BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:Europe/Helsinki BEGIN:DAYLIGHT DTSTART:20260329T040000 TZOFFSETFROM:+0200 TZOFFSETTO:+0300 RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3 TZNAME:EEST END:DAYLIGHT BEGIN:STANDARD DTSTART:20261025T030000 TZOFFSETFROM:+0300 TZOFFSETTO:+0200 RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10 TZNAME:EET END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20260509T094645Z UID:571A4D66-39F9-4920-B12F-176D4E111802 DTSTART;TZID=Europe/Helsinki:20260502T070000 DTEND;TZID=Europe/Helsinki:20260502T080000 DESCRIPTION:Abstract: Learning and controlling robots under system uncertai nties is a fundamental yet challenging problem in robotics. Although model -free methods combined with feedforward neural networks have proved to be a recipe for success in offline settings\, it’s been widely accepted tha t these approaches suffer from inherent limitations\, including the inabil ity to learn the Jacobian matrix\, reliance on pseudo-inverse operations\, and disregard for joint constraints in real-time applications. However\, for practical autonomous service robots in manufacturing and disaster reli ef\, it is crucial they can simultaneously estimate unknown structural par ameters and generate optimal motions online. In this talk I will describe our recent work that establishes a unified framework for synchronous learn ing and control\, bridging the gap between data-driven estimation and opti mization-based planning. I will present three integrated technologies esta blishing this new capability: a Kalman filter-based algorithm for simultan eous estimation of structural parameters and optimization indices in discr ete time\, an acceleration-level hybrid control scheme unifying position\, orientation\, and force regulation through quadratic programming without matrix inversion\, and a velocity-compensated gradient neural network with orthogonal projection that eliminates position errors and joint drift. At the core of these approaches is the fusion of neurodynamic optimization w ith real-time sensor feedback. Our methods outperform prior model-free app roaches on industrial robotic arms and multi-robot platforms\, demonstrati ng superior robustness against kinematic and dynamic uncertainties. I will also present preliminary results on distributed competitive collaboration using k-WTA networks and event-triggered planning.\n\nSpeaker(s): Long\, \n\nVirtual: https://events.vtools.ieee.org/m/557462 LOCATION:Virtual: https://events.vtools.ieee.org/m/557462 ORGANIZER:shuai.li@oulu.fi SEQUENCE:14 SUMMARY:Synchronous Learning and Planning Strategies for Robots with Optimi zation Algorithms URL;VALUE=URI:https://events.vtools.ieee.org/m/557462 X-ALT-DESC:Description: <br /><p style="language: zh-CN\; line-height: 150% \; text-align: justify\; text-justify: inter-ideograph\; direction: ltr\; unicode-bidi: embed\; mso-vertical-align-alt: auto\; mso-line-break-overri de: none\; word-break: normal\; punctuation-wrap: hanging\; margin: 0pt 0i n 0pt 0in\;"><span style="color: rgb(0\, 0\, 0)\;"><span style="font-size: 11pt\; font-family: 'Times New Roman'\; font-variant: normal\; text-trans form: none\; letter-spacing: 0pt\; font-weight: bold\; font-style: normal\ ; vertical-align: baseline\;">Abstract: </span><span style="font-size: 11p t\; font-family: 'Times New Roman'\; font-variant: normal\; text-transform : none\; letter-spacing: 0pt\; font-weight: normal\; font-style: normal\; vertical-align: baseline\;">Learning and controlling robots under system u ncertainties is a fundamental yet challenging problem in robotics. Althoug h model-free methods combined with feedforward neural networks have proved to be a recipe for success in offline settings\, it&rsquo\;s been widely accepted that these approaches suffer from inherent limitations\, includin g the inability to learn the Jacobian matrix\, reliance on pseudo-inverse operations\, and disregard for joint constraints in real-time applications . However\, for practical autonomous service robots in manufacturing and d isaster relief\, it is crucial they can simultaneously estimate unknown st ructural parameters and generate optimal motions online. In this talk I wi ll describe our recent work that establishes a unified framework for synch ronous learning and control\, bridging the gap between data-driven estimat ion and optimization-based planning. I will present three integrated techn ologies establishing this new capability: a Kalman filter-based algorithm for simultaneous estimation of structural parameters and optimization indi ces in discrete time\, an acceleration-level hybrid control scheme unifyin g position\, orientation\, and force regulation through quadratic programm ing without matrix inversion\, and a velocity-compensated gradient neural network with orthogonal projection that eliminates position errors and joi nt drift. At the core of these approaches is the fusion of neurodynamic op timization with real-time sensor feedback. Our methods outperform prior mo del-free approaches on industrial robotic arms and multi-robot platforms\, demonstrating superior robustness against kinematic and dynamic uncertain ties. I will also present preliminary results on distributed competitive c ollaboration using </span><span style="font-size: 11pt\; font-family: 'Tim es New Roman'\; font-variant: normal\; text-transform: none\; letter-spaci ng: 0pt\; font-weight: normal\; font-style: italic\; vertical-align: basel ine\;">k</span><span style="font-size: 11pt\; font-family: 'Times New Roma n'\; font-variant: normal\; text-transform: none\; letter-spacing: 0pt\; f ont-weight: normal\; font-style: normal\; vertical-align: baseline\;">-WTA networks and event-triggered planning.</span></span></p> END:VEVENT END:VCALENDAR