Give a Robot an Oscar? The Comments Are Asking Whether It Can Walk on Carpet

Thomas Wolf posted a robot video on X with the caption “give this robot an oscar.” In the video, the robot moves smoothly and really does have a bit of a performance feel. Original tweet

The comments didn’t follow the Oscar thread. Ironik Kly asked whether it could walk on carpet. Thomas replied with one word: “Yep.” That answer is more worth looking at than the movements in the video. Carpet is soft, slippery, and uneven; for a small quadruped robot, walking stably on carpet is not a default capability. Many demos look good on smooth floors but give themselves away on carpet.

Another question came from Tom Moor: Does it use an off-the-shelf LLM to drive its body? Thomas did not answer. This question is key. The movements in the video could come from a reinforcement learning policy, traditional control, or a language model involved in high-level decision-making, but you can’t tell from the video alone. For AI practitioners, connecting an LLM to a robot body is not hard; what’s hard is getting it to execute stably in the real physical world.

Vladimir Getselevich asked more directly: Can its movements be programmed or trained? Again, no reply was seen. A robot video and a robot platform that anyone can use are two different things. The former can be done through teleoperation, pre-scripted routines, or carefully selected clips; the latter requires documentation, APIs, and reproducible training pipelines. Someone in the comments mentioned @RemiFabreRobot and said his videos are the best-looking. Between looking good and being usable, there is still a long way to go.

Someone put this video together with the toy scene in Blade Runner. Johoshua said that scene is really about to happen.

The comparison is a bit exaggerated, but understandable. Once a robot has movement rhythm and “expressions,” people easily see it as a character. However, the logic by which demo videos spread and the logic by which products are implemented have never been the same. Videos need a few seconds of wonder; products need several hours of stability every day.

Back to that Oscar. Giving a robot a performance award is a joke. But those comments about carpet, LLMs, and programming are the key to whether it can go from a video into reality. A demo that can make people laugh is already not easy. What’s more worth watching is whether, after the laugh, it can keep doing work.