Overview
Embodied AI company behind the Spark-0 tactile robot manipulation model, the HumanTouch tactile glove dataset and the SparkArena benchmark. Spark-0's authors include researchers from Xspark AI, Tsinghua University and MMLab at the University of Hong Kong.
Reality check
Spark-0 tactile pretraining (company-claimed)
Spark-0 is pretrained with 1,000 hours of human tactile glove data (HumanTouch). With cameras only at test time, Ego Humanoid success rose from 52.5% to 71.7% vs vision-only pretraining.
Spark-0 with touch at deployment (company-claimed)
On SparkArena-Real, five tactile-conditioned real-robot tasks, success rose from 44.0% to 52.0% vs vision-only pretraining on the same data and compute.
Verified record
- Verified deployments
- None on file
- Active incidents
- None on file
Safety record
No incidents on record for Xspark AI.
Only active incidents are counted. Retracted incidents are excluded from this summary but remain reachable at their canonical URLs.
Full safety record: incidents, sourcing, and exposure data →
Recent activity
Every change to this record is dated, sourced, and independently verified where marked.
- Record createdOct 16, 2026
Added to the verified registry
Sources (2)
- Spark-0: Scaling Spatially Aligned Tactile Pretraining for Robot Dexterous Manipulation · https://xsparkai.com/Spark-0 · 2026-10-16
- Spark-0: Scaling Spatially Aligned Tactile Pretraining for Robot Dexterous Manipulation · https://arxiv.org/abs/2610.22342 · 2026-10-16
Common questions
What is Xspark AI?
What does Xspark AI make?
Is Xspark AI publicly traded?
How can I invest in Xspark AI?
Who owns Xspark AI?
Is Xspark AI a top robotics company?
Is Xspark AI safe?
Methodology: Verified · 2 sources (no primary) · last reviewed 2026-10-31
Verification posture
Verified
Low confidence
Review state
Stable
Last reviewed 2026-10-31
Sources by quality tier
- 1
- unclassified
- Unclassified source
- 1
- preprint
- Preprint
The framework is documented at /methodology. Corrections at /corrections. Reviewer: DEPLOY editorial team.
Methodology surface for Xspark AI.Machine-readable surfaces
- Markdown mirror: /companies/xspark-ai.md
- RSS feed: /companies/xspark-ai/feed.xml
- JSON-LD: embedded in this page’s head
- REST API: /v1/companies/72c252b1-065e-4e65-8c2f-3021e29c0ef0
- Revision history: /companies/xspark-ai/history
- Data documentation: /data
- Query this programmatically: Deploy MCP
Reality vs attention
Not enough verified signal yet to place Xspark AI against peers. Reality and attention percentiles publish once the underlying record clears its data floor.
6-month trend
Analysis
Limited public visibility and press coverage relative to peers.
Dimension breakdown
Verified signal
Attention (reach, not merit)
DEPLOY Intelligence scores are computed from verified registry data: confirmed deployments, disclosed funding rounds, regulatory filings, active job listings, video viewership, and press coverage. Confidence ratings reflect data availability. Scores update nightly.
DEPLOY Indices — verified vs claimed
Last computed: Oct 29, 2026