Emtorch experiments orchestrator for embedded systems

When executing experiments or tests on embedded environments, one often faces the challenge of performing multiple tasks in a repeatable and observable manner. For example: reset board, ensure embedded software booted, send trigger data using a selected link, monitor peripheral state to detect changes in behaviour, etc.

This is what emtorch helps to orchestrate: it runs various tools and scripts in a specific manner, then gathers their results for further inspection.

emtorch (previously known as emfuzzer, renamed in v2.0.0) is developed at the Warsaw University of Technology and licensed under the MIT License.

Installation

Emtorch requires Python 3.14 or later and is available on PyPI. It is recommended to install it in an isolated environment using Python venv, pipx, or uv.

Using venv:

python -m venv .venv
source .venv/bin/activate
pip install emtorch

Using pipx:

pipx install emtorch

Using uv:

uv tool install emtorch

Usage

To run experiments, use the run subcommand:

emtorch run --config=experiment.toml test1.bin test2.bin

For each specified data file, steps from experiment.toml will be executed and gathered results stored in a file named emtorch-CURRENTDATE.json. Application logs are output to the console and also stored in a .log file next to the .json results file. The prefix for output files can be modified using the --output-prefix option.

See default-config.toml in the source directory for a comprehensive example of experiment definition (this file can be safely used - the experiment calls cat on each passed file).

To obtain complete command line documentation, call:

emtorch --help
emtorch run --help

Additional options include:

• --repeats N - repeat each test case N times
• --repeat-mode {aabb,abab} - control repetition order

Quick Start

Create a simple test data file:

echo "Hello from test case" > test_data.txt

Create a minimal configuration file (config.toml):

[delays]
between_cases = 0.2
before_actions = 0.0

[[actions]]
type = "shell"
name = "read_data"

[actions.args]
cmd = "cat $EMTORCH_DATA_PATH"

Run your first experiment:

emtorch run test_data.txt --config config.toml

This will execute the configuration against the test data and output results in JSON format.

For a comprehensive configuration example, see default-config.toml in the repository.

Experiment Lifecycle

Each data file passed to emtorch represents a single Test Case. For each test case, the following experiment steps are performed:

1. Setup tasks are executed sequentially and their results stored.
2. Monitoring tasks are started (run concurrently in background).
3. Delay before actions (if configured).
4. Case actions are performed sequentially and their results stored.
5. Monitoring tasks finish when actions complete, results stored.
6. Check tasks are executed sequentially and their results stored.
7. Delay between cases (if more test cases remain).
8. Go to step 1 for the next Test Case.

Note: Failure of setup tasks does not interrupt test case execution - it is logged and stored in results, and subsequent steps are still executed for later analysis.

Configuration

Experiment configuration is stored in TOML format.

The configuration file defines four types of subtasks:

• setups - pre-case configuration tasks (sequential)
• monitoring - background observation tasks (concurrent)
• actions - main experiment operations (sequential)
• checks - post-case verification tasks (sequential)

Example configuration structure:

[delays]
between_cases = 0.2
before_actions = 1.0

[[setups]]
type = "ping-alive"
name = "check_host"

[setups.args]
host = "192.168.1.100"
timeout = 10
interval = 1

[[actions]]
type = "shell"
name = "run_test"

[actions.delays]
before = 0.5
after = 0.1

[actions.args]
cmd = "cat $EMTORCH_DATA_PATH"

[[checks]]
type = "ping-stable"
name = "verify_host"

[checks.args]
host = "192.168.1.100"
count = 3
interval = 1

Template Variables

Configuration values support $-string interpolation using template variables. Both $KEYWORD and ${KEYWORD} syntax are supported. Use $$ to escape the $ character.

Available template variables:

• $EMTORCH_CASE_ID - unique identifier of the current case
• $EMTORCH_DATA_PATH - full path to the case data file
• $EMTORCH_DATA_FILENAME - filename only of the case data

Additional variables can be introduced using --map argument.

Example usage:

[[actions]]
type = "shell"
name = "process"

[actions.args]
cmd = "process_data --input $EMTORCH_DATA_PATH --id $EMTORCH_CASE_ID"

SubTasks

Subtasks are the building blocks of experiments. Each subtask has a type, name, and type-specific arguments.

Available subtask types:

Subtask	Purpose
echo	Print messages to logs
exec	Execute programs with arguments
shell	Execute shell commands
remote	Execute commands on remote hosts via SSH
ping-alive	Check network connectivity (flood ping until first response)
ping-stable	Verify stable network response (all pings must succeed)
sftp-get	Download files from remote hosts
sftp-put	Upload files to remote hosts
file-write	Write content to local files
logger-int-matcher	Extract integer values from logs using regex
logger-float-matcher	Extract float values from logs using regex
coap-monitor	Monitor CoAP protocol messages
coap-send	Send CoAP protocol messages

For detailed subtask documentation:

• Run emtorch subtasks to list all available subtasks
• Run emtorch subtask <NAME> to see documentation for a specific subtask

Project Information

• License: MIT License
• Institution: Warsaw University of Technology
• GitHub: https://github.com/ZBOSK-II/emtorch
• Changelog: https://github.com/ZBOSK-II/emtorch/blob/master/CHANGELOG.md
• Previous name: emfuzzer (renamed in v2.0.0)