楊辰光，英國(guó)利物浦大學(xué)機(jī)器人學(xué)講席教授，機(jī)器人及自主系統(tǒng)研究團(tuán)隊(duì)帶頭人，歐洲科學(xué)與藝術(shù)院院士，國(guó)際電氣與電子工程師協(xié)會(huì)（IEEE）、英國(guó)工程技術(shù)學(xué)會(huì)（IET）、英國(guó)機(jī)械工程師學(xué)會(huì)（IMechE）、亞太人工智能學(xué)會(huì)（AAIA）及英國(guó)計(jì)算機(jī)學(xué)會(huì)（BCS）會(huì)士。現(xiàn)任IEEE柔性制造協(xié)同自動(dòng)化技術(shù)委員會(huì)（CAFM）聯(lián)席主席，《Robot Learning》創(chuàng)刊主編、《IEEE系統(tǒng)、人與控制論匯刊：系統(tǒng)》及《IEEE自動(dòng)化科學(xué)與工程匯刊》高級(jí)編輯，《Frontiers in Robotics and AI》期刊機(jī)器人計(jì)算智能領(lǐng)域首席主編。曾以大會(huì)主席身份成功籌辦第25屆IEEE工業(yè)技術(shù)國(guó)際會(huì)議（ICIT）和第27屆自動(dòng)化與計(jì)算國(guó)際會(huì)議（ICAC）。先后獲得2012年“IEEE機(jī)器人學(xué)匯刊最佳論文獎(jiǎng)”及2022年“IEEE神經(jīng)網(wǎng)絡(luò)與學(xué)習(xí)系統(tǒng)匯刊杰出論文獎(jiǎng)”兩大國(guó)際頂級(jí)期刊獎(jiǎng)項(xiàng)。
報(bào)告摘要：
Learning from Demonstration (LfD), or imitation learning, allows robots to acquire and generalize task skills through human demonstrations, creating a seamless integration of artificial intelligence and robotics. Most LfD approaches often overlook the importance of demonstrated forces and rely on manually configured impedance parameters. In response, my team has developed a series of biomimetic impedance and force controllers inspired by neuroscientific findings on motor control mechanisms in humans, enabling robots to imitate compliant manipulation skills. Our models reduce the dimensionality of skill representation, facilitating online optimization and reducing system sensitivity to parameter changes. To improve robot skill learning through enhanced perceptual capabilities, we designed anthropomorphic visual tactile sensors that assess contact force, surface texture, and shape, closely resembling the softness and wear resistance of human fingers for superior manipulation. The control and learning technologies we have developed have been particularly effective in robot teleoperation and human-robot collaboration, with shared control-based semi-autonomous methods that effectively integrate human intent with robotic autonomy, thereby achieving greater efficiency and usability.
