Notifications
Clear all
Exigo Trading Post
1
Posts
1
Users
0
Reactions
6
Views
Topic starter
How does a robotic or prosthetic hand master a complex task like grasping and rotating a ball? The difficulty lies in coordinating finger movements to apply just the right forces to an object—a task that has long been attributed to the sensitive skin and nerve endings in our hands. Inspired by this challenge, researchers at the ValeroLab in USC’s Viterbi School of Engineering set out to question a key assumption in robotics: Is tactile sensation truly essential for learning to manipulate objects?
In a new study published in Science Advances, researchers Romina Mir, Ali Marjaninejad, Andrew Erwin, and Professor Francisco Valero-Cuevas, along with collaborators from the University of California, Santa Cruz (UCSC), explore how a robotic hand learns through a combination of nature (built-in sensors) and nurture (the way it is trained). Their findings challenge long-held beliefs in the field.
The paper, titled “Curriculum Is More Influential Than Haptic Information During Reinforcement Learning of Object Manipulation Against Gravity,” reveals that the learning process—or “curriculum”—plays a more critical role in skill acquisition than the presence of tactile sensors.
https://github.com/ChrisHNE/kbzte39
https://github.com/DavidKEP9/Kbt934
https://github.com/PaulKBT/Kpt834
https://github.com/JeffRBt/Vrelk78
https://github.com/ChrisDNT9/pkdl9
https://github.com/DannyYAT/Pltr45
https://github.com/PeterKBN/Pkt9
https://github.com/CodyBLT/Dter46
https://github.com/DanielOBT/rxtd8
https://github.com/SteveWRB/kpfd9
https://github.com/RyanGSTR/plkt5
https://github.com/JeffRBT8/pkts59
https://github.com/MichaelBRTG/onkd5
https://github.com/RichardKVT/rkt5
https://github.com/CodyTNN/eklt5
https://github.com/NathanGKT/rcas5
https://github.com/TravisKNT/pkts
https://github.com/SteveTSK9/pkx5
https://github.com/BradleyEGT/ctsk
https://github.com/JoshGBT/HOCR
Using computational modeling and reinforcement learning, the team simulated a three-fingered robotic hand. Their goal was to test whether the hand could learn to manipulate an object against gravity even with limited or no tactile input. The answer: yes, it can—if trained in the right way.
“The sequence in which learning happens is crucial,” said Valero-Cuevas, who also holds an appointment in the Division of Biokinesiology and Physical Therapy at USC. “Just as biological organisms learn from their experiences, artificial systems benefit tremendously from how and when feedback is delivered during training.”
Romina Mir, a Ph.D. student and co-lead author, emphasized that the study provides a counterpoint to the traditional view. “We show that with the right reward structure and training sequence, a robotic hand can still learn complex manipulations—even without full tactile sensation,” she explained.
The research highlights a key insight: while tactile feedback can be beneficial, it is not always a prerequisite for learning effective motor control. Instead, designing an effective learning sequence may have an even greater impact on a robotic system’s ability to adapt and perform in the physical world.
This interdisciplinary collaboration between USC and UCSC included co-lead author Parmita Ojaghi and Professor Michael Wehner from UCSC, with additional contributions from Ali Marjaninejad and Andrew Erwin at USC.
The findings could have wide-reaching implications not only for the development of more efficient robotic systems and prosthetics but also for advancing artificial intelligence that better mirrors how humans and animals learn through experience.
Posted : 04/04/2025 10:00 am
How does a robotic or prosthetic hand master a complex task like grasping and rotating a ball? The difficulty lies in coordinating finger movements to apply just the right forces to an object—a task that has long been attributed to the sensitive skin and nerve endings in our hands. Inspired by this challenge, researchers at the ValeroLab in USC’s Viterbi School of Engineering set out to question a key assumption in robotics: Is tactile sensation truly essential for learning to manipulate objects?
In a new study published in Science Advances, researchers Romina Mir, Ali Marjaninejad, Andrew Erwin, and Professor Francisco Valero-Cuevas, along with collaborators from the University of California, Santa Cruz (UCSC), explore how a robotic hand learns through a combination of nature (built-in sensors) and nurture (the way it is trained). Their findings challenge long-held beliefs in the field.
The paper, titled “Curriculum Is More Influential Than Haptic Information During Reinforcement Learning of Object Manipulation Against Gravity,” reveals that the learning process—or “curriculum”—plays a more critical role in skill acquisition than the presence of tactile sensors.
https://github.com/ChrisHNE/kbzte39
https://github.com/DavidKEP9/Kbt934
https://github.com/PaulKBT/Kpt834
https://github.com/JeffRBt/Vrelk78
https://github.com/ChrisDNT9/pkdl9
https://github.com/DannyYAT/Pltr45
https://github.com/PeterKBN/Pkt9
https://github.com/CodyBLT/Dter46
https://github.com/DanielOBT/rxtd8
https://github.com/SteveWRB/kpfd9
https://github.com/RyanGSTR/plkt5
https://github.com/JeffRBT8/pkts59
https://github.com/MichaelBRTG/onkd5
https://github.com/RichardKVT/rkt5
https://github.com/CodyTNN/eklt5
https://github.com/NathanGKT/rcas5
https://github.com/TravisKNT/pkts
https://github.com/SteveTSK9/pkx5
https://github.com/BradleyEGT/ctsk
https://github.com/JoshGBT/HOCR
Using computational modeling and reinforcement learning, the team simulated a three-fingered robotic hand. Their goal was to test whether the hand could learn to manipulate an object against gravity even with limited or no tactile input. The answer: yes, it can—if trained in the right way.
“The sequence in which learning happens is crucial,” said Valero-Cuevas, who also holds an appointment in the Division of Biokinesiology and Physical Therapy at USC. “Just as biological organisms learn from their experiences, artificial systems benefit tremendously from how and when feedback is delivered during training.”
Romina Mir, a Ph.D. student and co-lead author, emphasized that the study provides a counterpoint to the traditional view. “We show that with the right reward structure and training sequence, a robotic hand can still learn complex manipulations—even without full tactile sensation,” she explained.
The research highlights a key insight: while tactile feedback can be beneficial, it is not always a prerequisite for learning effective motor control. Instead, designing an effective learning sequence may have an even greater impact on a robotic system’s ability to adapt and perform in the physical world.
This interdisciplinary collaboration between USC and UCSC included co-lead author Parmita Ojaghi and Professor Michael Wehner from UCSC, with additional contributions from Ali Marjaninejad and Andrew Erwin at USC.
The findings could have wide-reaching implications not only for the development of more efficient robotic systems and prosthetics but also for advancing artificial intelligence that better mirrors how humans and animals learn through experience.
