Implementation:Haosulab ManiSkill BaseSensor

Knowledge Sources	Haosulab_ManiSkill
Domains	Robotics, Simulation, Sensors
Last Updated	2026-02-15 08:00 GMT

Overview

Concrete tool for defining the abstract base class interface for all sensors in ManiSkill.

Description

The base_sensor.py module defines two foundational types:

BaseSensorConfig -- A minimal dataclass holding sensor configuration:

uid: str -- Unique identifier for the sensor.

BaseSensor -- Abstract base class that all ManiSkill sensors must implement:

__init__(config) -- Stores the sensor configuration.
setup() -- Called during environment/scene reconfiguration to initialize the sensor for the current scene.
capture() -- Captures sensor data. Designed to be non-blocking where possible (e.g., taking a picture after scene render update).
get_obs(**kwargs) -- Returns the captured data as an observation dictionary for agent consumption. Must be implemented by subclasses.
get_params() -- Returns sensor parameters as a dictionary of torch tensors. Must be implemented by subclasses.
get_images() -- Returns sensor data visualized as an image (shape B, H, W, 3). Not for agent observations. Must be implemented by subclasses.
uid property -- Returns the sensor's unique identifier from its config.

Usage

Subclassed by all sensor implementations (Camera, StereoDepthCamera, etc.). Use when implementing custom sensors for ManiSkill environments.

Code Reference

Source Location

Repository: Haosulab_ManiSkill
File: mani_skill/sensors/base_sensor.py

Signature

@dataclass
class BaseSensorConfig:
    uid: str

class BaseSensor:
    def __init__(self, config: BaseSensorConfig) -> None: ...
    def setup(self) -> None: ...
    def capture(self) -> None: ...
    def get_obs(self, **kwargs): ...      # abstract
    def get_params(self) -> dict: ...     # abstract
    def get_images(self) -> torch.Tensor: ...  # abstract
    @property
    def uid(self) -> str: ...

Import

from mani_skill.sensors.base_sensor import BaseSensor, BaseSensorConfig

I/O Contract

Inputs

Name	Type	Required	Description
config	BaseSensorConfig	Yes	Configuration with at minimum a unique identifier

Outputs

Name	Type	Description
obs	dict	Observation data from get_obs()
params	dict	Sensor parameters (torch tensors) from get_params()
images	torch.Tensor	Visualization images of shape (B, H, W, 3)

Usage Examples

Basic Usage

from mani_skill.sensors.base_sensor import BaseSensor, BaseSensorConfig

class CustomSensor(BaseSensor):
    def get_obs(self, **kwargs):
        return {"custom_data": self.captured_data}

    def get_params(self) -> dict:
        return {"param1": torch.tensor([1.0])}

    def get_images(self):
        return self.visualization_image

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment