Sony researchers enhance Aibo’s existing features using AI

[Chris]

Huge news! Sophisticated sim-to-real based reinforcement learning methods have been employed by Sony researchers to enhance Aibo’s existing features.

The researchers have enabled Aibo to achieve an even quieter walking ability using AI in a simulated environment.

They have also enabled Aibo to dance more realistically to music with a greater range of smoother motion while maintaining eye contact with the user all while responding to dynamic movements.

This is a massive Aibo development! The exciting question is will Sony be implementing these updated features in a future software update? Or perhaps a more exciting possibility, will Sony add completely new features developed with the help of AI?!

DFM: Deep Fourier Mimic for Expressive Dance Motion Learning

DFM: Deep Fourier Mimic for Expressive Dance Motion Learning

As entertainment robots gain popularity, the demand for natural and expressive motion, particularly in dancing, continues to rise. Traditionally, dancing motions have been manually designed by artists, a process that is both labor-intensive and restricted to simple motion playback, lacking the...

arxiv.org

 

[Chris]

“Sony Tackles Walking Noise Issue in Bipedal Robots with AI

A major problem for humanoid robots in the future will be "walking noise." While this issue has not yet gained attention in the US or China, where humanoid robot development is booming, Sony has developed a technology that could address this problem.

In collaboration with the Swiss Federal Institute of Technology (ETH Zurich), Sony has developed a technology that uses reinforcement learning to reduce the walking noise of its robotic dog "aibo" without modifying its hardware. This is the first time reinforcement learning has been used to reduce walking noise in robots.

The technology uses a "Sim-to-Real" approach, where a policy learned in a simulation environment is transferred to a real-world robot. Sony used NVIDIA's Isaac robot development platform for the simulation.

Sony's efforts to reduce aibo's walking noise were prompted by customer feedback. The company has been working on this issue for the past year or two and has already updated aibo's firmware to reduce the noise. The next step was to use reinforcement learning to further reduce the noise.

However, Sony found that traditional reinforcement learning approaches can actually increase walking noise, as the robot learns to walk with stronger foot strikes to avoid falling. To address this issue, Sony developed a new approach that uses reinforcement learning to specifically reduce walking noise.

The technology is not only applicable to small bipedal robots like aibo but also to larger humanoid robots. According to Sony, the technology is particularly useful for bipedal robots, which have a harder time maintaining balance than quadruped robots and are more prone to increased walking noise.”

Source: https://xtech.nikkei.com/atcl/nxt/column/18/03118/00005/