← Back to videos
A robot once delivered a charger to my hotel room in China. MAX from @Nullmax_ai is what that idea

Shruti
@heyshrutimishra
November 13, 2026· 3:22
A robot once delivered a charger to my hotel room in China. MAX from @Nullmax_ai is what that idea looks like with a brain. It's a robot dog running NPI-Brain. Tasks go in by voice, and it handles being interrupted mid-sentence. The part I like most is the memory. It knows the meeting room, reception and elevator lobby by name, and learns from its past runs. It also navigates across floors, recognizes people and replans when something blocks its path. For the next clip, I'd love to see what other real-world scenarios Nullmax will bring MAX into next. #EmbodiedAI #NPIBrain
Transcript
- Parkour, cross obstacles
- 0:02
- pull-off backflips—today's robot dogs have already maxed out
- 0:05
- their athletic prowess
- 0:07
- But can they break free from pre-programmed routines
- 0:10
- understand natural language, and autonomously execute sequential tasks?
- 0:14
- This is the critical threshold for embodied intelligence
- 0:18
- to become reality
- 0:19
- As a leading physical AI company
- 0:22
- NullMax extends real-world validated AI capabilities into the
- 0:25
- broad physical realm
- 0:27
- It has fully self-developed a general physical AI technology stack
- 0:30
- NullMax Physical AI, and built the embodied brain
- 0:34
- NPI Brain Atopic
- 0:35
- The platform delivers end-to-end capabilities spanning perception
- 0:39
- comprehension, memory, decision-making, motion, and manipulation control
- 0:45
- enabling machines to evolve from seeing and understanding
- 0:48
- to interacting with the world autonomously
- 0:52
- No remote control, no pre-planned path, just one verbal instruction
- 0:56
- and we see whether it can wrap up
- 0:58
- the whole workflow on its own
- 1:02
- Look
- 1:03
- no predefined route or pre-built map is supplied
- 1:05
- for this mission
- 1:06
- It draws on its own visual memory
- 1:09
- Reaching an intersection, it pauses to orient itself
- 1:13
- I recognize this photo wall
- 1:15
- turn right here
- 1:16
- I also remember this parcel locker
- 1:18
- turn left here
- 1:20
- This is NPI's tokenized scene memory
- 1:22
- What it sees, hears
- 1:24
- and its own state are all stored within its contextual memory
- 1:27
- independent of conventional SLAM mapping
- 1:30
- When its original path is blocked by a person
- 1:32
- it does not force its way through
- 1:34
- Instead
- 1:35
- it detours via the sofa area
- 1:37
- before resuming its route toward the target
- 1:40
- Upon entering the meeting room
- 1:42
- it cross-references fascia profile photos
- 1:45
- one by one to locate the person who
- 1:47
- gave it the task
- 1:48
- 去落叶天地大
- 1:49
- 王茶祭门口找
- 1:56
- 收件人送
- 2:01
- 过
- 2:07
- 文件飞书和手机
- 2:25
- 号已经发给你了
- 2:41
- The moment the recipient waves
- 2:43
- the robot identifies them instantly
- 2:45
- From parsing the command to completing delivery
- 2:48
- the whole process runs with zero human intervention
- 2:51
- This on-device brain ties together four core abilities:
- 2:55
- conversing, navigating, remembering, and getting tasks done
- 2:59
- To be clear, this is still testing within controlled environments
- 3:03
- and widespread consumer adoption remains down the line
- 3:06
- Still, the trajectory is clear:
- 3:08
- whether a quadruped robot can do a backflip
- 3:10
- is no longer the main draw
- 3:12
- What matters
- 3:13
- is whether it can autonomously carry out complex
- 3:15
- human-assigned tasks without human oversight
Watch these videos on your TV
XTTV brings video content from X to your living room. Download now and start watching on the big screen.