Robot Runtime Observability
How can a robot run retain its state, versions, environment, and task context?
Connect topics, TF, logs, parameters, maps, and task events into run records that can be queried and compared.
INDEPENDENT ROBOTICS RESEARCH
Understand how robots run.Understand why they fail.
Zephastra Robotics studies robot runtime, diagnosis, and reliability through reproducible experiments, verifiable code, and public research notes while gradually building ZephaRobot.
RESEARCH QUESTION
How can a robot failure become a replayable, explainable, and verifiable engineering record?
01
Robot Runtime
Topic · TF · Sensor · Log
02
Episode
Mission · Timeline · Context
03
Diagnosis
Incident · Rule · Evidence
04
Review
Replay · Report · Decision
Research agenda
Three long-term research themes.
Failure Replay & Diagnosis
How can fragmented data reconstruct a failure timeline without losing the original evidence?
Study reproducible and explainable incident review through Episodes, Incidents, and Evidence.
Navigation Reliability
How do TF, odometry, localization, planning, and control combine to produce navigation failures?
Build a body of SLAM, Nav2, and sensor-chain failure cases around ROS 2 differential-drive robots.
ZephaRobot
Turn a robot run into a reviewable Episode.
ZephaRobot is an early research prototype that connects missions, timelines, signals, incidents, and evidence. The current implementation validates this workflow with mock and simulated data.
Robot Runtime
Topic · TF · Sensor · Log
Episode
Mission · Timeline · Context
Diagnosis
Incident · Rule · Evidence
Review
Replay · Report · Decision
Engineering evidence
Research should leave inspectable evidence.
The direction is evaluated through code, tests, data structures, and reviewable reports—not product claims.
Episode data model
The current prototype includes server-side Robot, Mission, Episode, Signal, Event, Incident, and Report models and APIs.
Repeatable mock scenarios
Generated run data validates event timelines, incident rules, and the reporting workflow.
Explainable incident rules
Diagnosis summaries retain the triggering rule, time range, and related signals instead of returning an untraceable conclusion.
Episode review reports
Run summaries, incidents, evidence, and follow-up checks are assembled into a consistent HTML report.
Engineering experiments
From simulation and data to diagnostic tools.
Core research prototype
RESEARCHZephaRobot
Organizes a robot mission as a traceable Episode that connects timelines, runtime signals, incidents, and diagnostic evidence.
EpisodeDiagnosticsFastAPINext.js
View project Reproducible experiment environment
RESEARCHNavigation Failure Lab
Uses ROS 2, Gazebo, and Nav2 to construct repeatable navigation tasks and failure scenarios for diagnostic validation.
ROS 2GazeboNav2Fault Injection
View project Public engineering data asset
RESEARCHRobot Run Dataset
Builds replayable robot-run cases from rosbag2, logs, maps, parameters, and version metadata.
rosbag2MCAPTFReproducibility
View project Research journal
Problems, evidence, decisions, and results.
Public notes retain assumptions, failure observations, analysis, and validation methods.
OPEN RESEARCH & COLLABORATION
Start with a concrete robot problem.
Open to technical exchange around ROS 2 run records, Nav2 failures, sanitized data, reproducible simulation, and robot software engineering.
Contact