Too bad Ford stopped the UBI trial in Ontario.
Robotics and humanoid robotics & civilian drones discussion
- Thread starter tphuang
- Start date
Unitree humanoid robots are fully sold out, but don't worry, you can rent them for 10k RMB/day.
robotic hand with 15 DOFs. This looks more advanced than anything I've seen.
Looks like BYD is partnering up with Unitree. Previously, we've only known about Ubtech, Siasun & Fourier I think. This would be a huge contract for Unitree
The world's first universal embodied base model Genie Operator-1 (GO-1) was officially released. What's so great about this big model?
▼GO-1's core innovation lies in the "ViLLA architecture" (Vision-Language-Latent-Action), which deeply integrates the multimodal large model (VLM), hybrid expert system (MoE) and implicit action planning:
¹VLM: Through training with massive Internet graphic data, robots are given the ability to "see the world" and "understand language", such as recognizing cups of different shapes and understanding the "pour water" command. ²MoE-Latent Planner: Using cross-ontology and human operation video data, the action logic is refined, allowing robots to "think about steps" like humans, such as abstracting the general strategy of "move-wipe-avoid obstacles" from the video of wiping the table.
³MoE-Action Expert: Relying on millions of real machine data training, it can achieve millimeter-level precision operations, such as pouring water without spilling, and adaptively adjusting the strength when grabbing fragile items.
With the collaboration of the three, GO-1's success rate in tasks such as pouring water and cleaning is 32% higher than that of existing models, and its small sample learning ability breaks through the data bottleneck - only a small number of teaching videos are needed to generalize new skills.
▼GO-1's core innovation lies in the "ViLLA architecture" (Vision-Language-Latent-Action), which deeply integrates the multimodal large model (VLM), hybrid expert system (MoE) and implicit action planning:
¹VLM: Through training with massive Internet graphic data, robots are given the ability to "see the world" and "understand language", such as recognizing cups of different shapes and understanding the "pour water" command. ²MoE-Latent Planner: Using cross-ontology and human operation video data, the action logic is refined, allowing robots to "think about steps" like humans, such as abstracting the general strategy of "move-wipe-avoid obstacles" from the video of wiping the table.
³MoE-Action Expert: Relying on millions of real machine data training, it can achieve millimeter-level precision operations, such as pouring water without spilling, and adaptively adjusting the strength when grabbing fragile items.
With the collaboration of the three, GO-1's success rate in tasks such as pouring water and cleaning is 32% higher than that of existing models, and its small sample learning ability breaks through the data bottleneck - only a small number of teaching videos are needed to generalize new skills.
Last edited:
more from Unitree robot dogs and its products are put on sale for overseas market through AliExpress
BYD will announce something it looks like with Unitree robot on 21st
Zhiyuan Robotics released its latest bipedal humanoid robot, Lingxi X2.
It is reported that Lingxi X2 robot has extremely high athletic ability and can dance, run, ride a scooter, play a balance car, and even ride a bicycle.
In addition to its excellent athletic ability, Lingxi X2 also introduces the multimodal interactive large model "Silicon Photonics", which enables it to interact with humans more naturally and realistically, with millisecond-level interactive response.
It is reported that Lingxi X2 robot has extremely high athletic ability and can dance, run, ride a scooter, play a balance car, and even ride a bicycle.
In addition to its excellent athletic ability, Lingxi X2 also introduces the multimodal interactive large model "Silicon Photonics", which enables it to interact with humans more naturally and realistically, with millisecond-level interactive response.
Last edited:
MagicLab also released a demo of walking on straight knees