Understanding Codelets

Components are the basic building blocks of a robotics application. Isaac SDK includes various components which can use in your application. This tutorial uses an existing component to explain the various important concepts around components.

The public interface of the component DifferentialBaseOdometry, contained in the //packages/navigation, directory, is shown below. This component listens to the wheel odometry from a differential base and tries to estimate the pose of the robot.

Copy
Copied!
            

namespace isaac { namespace navigation { // Integrates (2D) odometry for a differential base to estimate it's // ego motion. class DifferentialBaseOdometry : public alice::Codelet { public: void start() override; void tick() override; // Incoming current dynamic state of the differential base which is used to estimate its // ego motion in an odometry frame (type: DifferentialBaseDynamics) ISAAC_PROTO_RX(StateProto, state); // Outgoing ego motion estimate for the differential base. ISAAC_PROTO_TX(Odometry2Proto, odometry); // Maximum acceleration to use (helps with noisy data or wrong data // from simulation) ISAAC_PARAM(double, max_acceleration, 5.0); // The name of the source coordinate frame under which to publish // the pose estimate. ISAAC_PARAM(std::string, odometry_frame, "odom"); // The name of the target coordinate frame under which to publish // the pose estimate. ISAAC_PARAM(std::string, robot_frame, "robot"); // 1 sigma of noise used for prediction model in the following order: // pos_x, pos_y, heading, speed, angular_speed, acceleration ISAAC_PARAM(Vector6d, prediction_noise_stddev, \ (MakeVector<double, 6>({0.05, 0.05, 0.35, 0.05, 1.00, 3.00}))); // 1 sigma of noise used for observation model in the following order: // speed, angular_speed, acceleration ISAAC_PARAM(Vector3d, observation_noise_stddev, \ (Vector3d{0.25, 0.45, 2.0})); // This is the pose under which the ego motion estimation will be // written to the pose tree. ISAAC_POSE2(odom, robot); private: ... }; } // namespace navigation } // namespace isaac ISAAC_ALICE_REGISTER_CODELET(isaac::navigation::DifferentialBaseOdometry);

The following sections step through DifferentialBaseOdometry, explaining each section.

Copy
Copied!
            

class DifferentialBaseOdometry : public alice::Codelet {

Codelets are very common components which enable you to write code which is executed repeatedly. DifferentialBaseOdometry is derived from the alice::Codelet.

Codelets run in one of the three following ways:

  • Tick periodically: The tick function is executed regularly after a fixed time period. A typical example is a controller which ticks 100 times a second to send control commands to hardware.
  • Tick on message: The tick function is executed whenever a new message is received. A typical example is an image processing algorithm which computes certain information on every new camera image which is captured.
  • Tick blocking: The tick function is executed immediately again after it has finished. A typical example is a hardware driver which reads on a socket in blocking mode.
Note

Always use blocking communication with hardware instead of threads, if possible. Isaac SDK automatically creates and manages the necessary threads.

When a codelet ticks, it prevents other codelets in the same node from ticking at the same time. To run codelets in parallel, place them in separate nodes.

DifferentialBaseOdometry uses a periodic tick. This is realized in the start function as shown below:

Copy
Copied!
            

void DifferentialBaseOdometry::start() { ... tickPeriodically(); ... }

The tick period itself is set in the configuration as explained later.

Many components receive or transmit messages to other components. Message passing is a powerful way to encapsulate components and ensuring modularity of the codebase.

Copy
Copied!
            

// Incoming current dynamic state of the differential base which is used to estimate its // ego motion in an odometry frame (type: DifferentialBaseDynamics) ISAAC_PROTO_RX(StateProto, state);

The ISAAC_PROTO_RX macro is used to define a receiving (RX) channel. The macros take two arguments: the type of the message and the name of the channel. Isaac SDK is not particularly tied to a specific message format, but currently cap’n’proto is in widespread use. For more information see the cap’n’proto website.

A message can be read on an receiving channel for example, as follows:

Copy
Copied!
            

const auto& rmp_reader = rx_state().getProto(); ... state_.speed() = rmp_reader.getLinearSpeed();

The function rx_state is automatically generated by the ISAAC_PROTO_RX macro, and a StateProto message containing a DifferentialBaseDynamics is expected. All message schemas of the Isaac SDK can be found in the //message folder or in the corresponding section in this document.

At the end of the tick, after all computations are done, a component often wants to send out a new message to whomever is listening.

Copy
Copied!
            

// Outgoing ego motion estimate for the differential base. ISAAC_PROTO_TX(Odometry2Proto, odometry)

The ISAAC_PROTO_TX macro is used to define a transmitting (TX) channel. This is very similar to the way in which ISAAC_PROTO_RX macro works.

A message can be created and sent as shown below:

Copy
Copied!
            

auto odom_builder = tx_odometry().initProto(); ToProto(odom_T_robot, odom_builder.initOdomTRobot()); odom_builder.setSpeed(state_.speed()); ... tx_odometry().publish();

Again the tx_odometry function is automatically created by the ISAAC_PROTO_TX macro. Use initProto to start a new message on this channel. Functions automatically generated by the cap’n’proto schema like initOdomTRobot and setSpeed can be used to write data to the message proto. When the message is complete it can be send via the publish() function. Only one message can be generated at a time.

Primary data types such as integers are supported directly by cap’n’proto, but it can be more difficult to process complicated data types. To handle such cases, Isaac SDK provides convenient ToProto/FromProto functions, which have common patterns like following:

Copy
Copied!
            

// Writes a UUID to a proto void ToProto(const Uuid& uuid, ::UuidProto::Builder builder) // Reads a UUID from a proto Uuid FromProto(::UuidProto::Reader reader)

or

Copy
Copied!
            

// Parses a tensor from a message. This version parses tensors of any element type and rank. bool FromProto(::TensorProto::Reader reader, const std::vector<isaac::SharedBuffer>& buffers, isaac::UniversalTensorConstView<Storage>& universal_view); // Creates a tensor from a proto. Will print errors and return false if the tensor type is not // compatible with the proto. bool FromProto(::TensorProto::Reader reader, const std::vector<isaac::SharedBuffer>& buffers, isaac::TensorBase<K, Dimensions, BufferType>& tensor_view)

See the messages/ folder for header files containing ToProto/FromProto functions.

Complicated algorithms can often be parameterized in various different ways. ISAAC_PARAM allows you to define a configuration parameter which can be set via configuration, read in the code, and changed in the frontend.

Copy
Copied!
            

// Maximum acceleration to use (helps with noisy data or wrong data // from simulation) ISAAC_PARAM(double, max_acceleration, 5.0)

There are three parameters to ISAAC_PARAM:

  • type: This is the type of configuration parameter. The basic types are int, double, bool and std::string. Isaac SDK also provides support for various math types like Pose2/3, SO2/3, and Eigen vectors and matrices. An STD vector of any of those types is also supported.
  • name: The name defines the key under which the parameter is stored in the configuration file and the function name under which it can be accessed in code.
  • default value: In case no value is specified in the configuration file this value is used instead. The default can also be omitted which forces the user to specify a value in the configuration file.

In the example of DifferentialBaseOdometry the tick function starts with retrieving the desired prediction noise used in the Kalman filter:

Copy
Copied!
            

void DifferentialBaseOdometry::tick() { navigation::DifferentialBaseState prediction_noise_stddev; prediction_noise_stddev.elements = get_prediction_noise_stddev();

Configuration can be changed in multiple ways:

  • The default configuration parameter can be changed. This should be used with caution, because it’ changes the value for all applications which have not overwritten the value in a configuration file.

  • The value can be set in a JSON configuration file. Most sample applications include a JSON file where various parameters are set. For example in //app/samples/simple_robot, a configuration parameter can be changed by adding the following JSON to the config section:

    Copy
    Copied!
                

    { "config": { ... "segway_odometry": { "isaac.navigation.DifferentialBaseOdometry": { "max_acceleration": 2.0 } } ... } }


In this example segway_odometry is the name of the node which contains a component of our type with name isaac.navigation.DifferentialBaseOdometry.

Every Isaac application is based on a JSON file. The JSON file describes the dependencies of the application, the node graph and the message flow, and contains custom configuration. The basic structure of a JSON application file:

Copy
Copied!
            

{ "name": "my_application", "modules": [ ... ], "graph": { "nodes": [ ... ], "edges": [ ... ] }, "config": { ... } }


The “name” given of the application must match the name given to the isaac_app in the appropriate BUILD file, and must also match the filename of the JSON file (in this case, “my_application.app.json”).

The “modules” list enumerates all the packages that contain components used in this application. The “modules” list specified for the isaac_app in the BUILD file must contain the list in the JSON.

An application “graph” defines “nodes” that correspond to the components in use, with “edges” connecting the nodes. The edges determine the message passing sequence between different nodes. An example of the “nodes” section of the app graph:

Copy
Copied!
            

"nodes": [ { "name": "node_1", "components": [ { "name": "message_ledger", "type": "isaac::alice::MessageLedger" }, { "name": "component_a", "type": "isaac::alice::ComponentX" } ] }, { "name": "node_2", "components": [ { "name": "message_ledger", "type": "isaac::alice::MessageLedger" }, { "name": "component_b", "type": "isaac::alice:CodeletY" } ] } ]


Note that the “type” parameter must match the name given to the component by the macro ISAAC_REGISTER_COMPONENT or ISAAC_REGISTER_CODELET. In addition, each node must contain a message_ledger component in order to handle the message passing to/from the node.

The edges determine the message passing sequence between different components. Each edge requires a source and a target. Using the example above, an edge between the “output” message of node_1’s component_a and the “input” message of node_2’s component_b looks like this:

Copy
Copied!
            

"edges": [ { "source": "node_1/component_a/output", "target": "node_2/component_b/input" }, ]


This example assumes component_a has an ISAAC_PROTO_TX message defined with the name “output” and component_b has an ISAAC_PROTO_RX message defined with the name “input” of the same type.

The application JSON file also contains configuration data, or “config”, for various parameters to customize behavior. Every configuration parameter is referenced by three elements: node name, component name, and parameter name.

Configuration of a floating point value parameter named “param” of node_1’s component_a looks like this:

Copy
Copied!
            

"config": { "node_1": { "component_a": { "param": 0.1 } } }


This assumes that component_a has defined an ISAAC_PARAM by the name of “param” and set it to 0.1.

See Developing Codelets in C++ for a tutorial that includes building an application graph using the concepts introduced above.

As more components are added to applications, application graphs can get lengthy and repetitive. Subgraphs can prove useful in simplifying application graphs. When using and re-using nodes with multiple components connected together, a JSON subgraph can include required components, edges, and configurations, so that you can add a relatively high-level group without concern for the lower-level details.

In the following diagram, nodes A, B, and C in App 1 and App 2 are identical. Instead of duplicating them in the JSON for each application, a subgraph X can be created.

subgraphs1.png

In this way, the graphs for App 1 and App 2 are greatly simplified, as shown below. This abstraction simplifies the apps, lowers the maintenance, hides the expertise, and provides a better user experience.

subgraphs2.png

An example of a subgraph is apps/carter/carter_hardware.subgraph.json, which contains edges and components needed for almost any application pertaining to Carter hardware. Rather than repeating that information in the JSON file for every application, the subgraph can be included in the larger JSON configuration for the entire application. Below is an example of the carter_hardware subgraph, which has nodes for Segway base, Velodyne lidar, and BMI160 inertial measurement unit (IMU). The component type isaac::alice::Subgraph makes input/output with subgraphs cleaner; an example is provided later in this section. In addition to a “graph” section, a subgraph also includes “modules”, “edges”, and “config” as illustrated below.

Copy
Copied!
            

{ "modules": [ "imu", "segway", "velodyne_lidar" ], "graph": { "nodes": [ { "name": "subgraph", "components": [ { "name": "message_ledger", "type": "isaac::alice::MessageLedger" }, { "name": "interface", "type": "isaac::alice::Subgraph" } ] }, { "name": "segway_rmp", "components": [ { "name": "message_ledger", "type": "isaac::alice::MessageLedger" }, { "name": "isaac.SegwayRmpDriver", "type": "isaac::SegwayRmpDriver" }, { "name": "isaac.alice.Failsafe", "type": "isaac::alice::Failsafe" } ] }, { "name": "vlp16_initializer", "components": [ { "name": "lidar_initializer", "type": "isaac::alice::PoseInitializer" } ] }, "name": "vlp16", "components": [ { "name": "message_ledger", "type": "isaac::alice::MessageLedger" }, { "name": "VelodyneLidar", "type": "isaac::velodyne_lidar::VelodyneLidar" } ] }, { "name": "imu", "components": [ { "name": "message_ledger", "type": "isaac::alice::MessageLedger" }, { "name": "IioBmi160", "type": "isaac::imu::IioBmi160" } ] } ], "edges": [ { "source": "subgraph/interface/diff_base_command", "target": "segway_rmp/isaac.SegwayRmpDriver/segway_cmd" }, { "source": "segway_rmp/isaac.SegwayRmpDriver/segway_state", "target": "subgraph/interface/diff_base_state" }, { "source": "vlp16/VelodyneLidar/scan", "target": "subgraph/interface/scan" }, { "source": "imu/IioBmi160/imu_raw", "target": "subgraph/interface/imu_raw" } ] }, "config": { "segway_rmp": { "isaac.SegwayRmpDriver": { "ip": "192.168.0.40", "tick_period": "20ms" }, "isaac.alice.Failsafe": { "name": "robot_failsafe" } }, "vlp16": { "VelodyneLidar": { "ip": "192.168.0.5" } }, "imu": { "IioBmi160": { "i2c_device_id": 1, "tick_period": "100Hz" } } } }

Employ subgraphs using the syntax shown below. It’s important to note that each node in a subgraph is prefixed with the subgraph name, e.g., carter1.segway_rmp. This allows configuring or creating edges with a specific instance of a subgraph.

Copy
Copied!
            

{ .... "graph": { "nodes": [ { "name": "carter1", "subgraph": "apps/carter/carter_hardware.subgraph.json" }, { "name": "carter2", "subgraph": "apps/carter/carter_hardware.subgraph.json" }, { "name": "imu_corrector", "components": [ { "name": "message_ledger", "type": "isaac::alice::MessageLedger" }, { "name": "ImuCorrector", "type": "isaac::imu::ImuCorrector" } ] }, ... ], "edges": [ { "source": "carter1.subgraph/interface/imu_raw", "target": "imu_corrector/ImuCorrector/raw" }, ... ] }, "config": { "imu_corrector": { "ImuCorrector": { "calibration_variance_stationary": 0.1, } }, "carter1.vlp16_initializer": { "lidar_initializer": { "pose": [1.0, 0.0, 0.0, 0.0, -0.04, 0.0, 0.59] } }, "carter2.vlp16_initializer": { "lidar_initializer": { "pose": [1.0, 0.0, 0.0, 0.0, -0.04, 0.0, 0.77] } }, ... } }

Note that when referring to the nodes in the carter_hardware subgraph, the name of the subgraph is used as prefix, e.g., carter1.subgraph, carter1.vlp16, or carter2.vlp16. Otherwise, the format is as explained in Application JSONs.

Subgraphs can be nested. For example, carter_hardware and scan_flattener subgraphs are used in 2d_carter.subgraph.json as shown below:

Copy
Copied!
            

{ "graph": { "nodes": [ { "name": "subgraph", "components": [ { "name": "message_ledger", "type": "isaac::alice::MessageLedger" }, { "name": "interface", "type": "isaac::alice::Subgraph" } ] }, { "name": "carter_hardware", "subgraph": "apps/carter/carter_hardware.subgraph.json" }, { "name": "scan_flattener", "subgraph": "packages/navigation/apps/scan_flattener.subgraph.json" } ], "edges": [ { "source": "carter_hardware.subgraph/interface/imu_raw", "target": "subgraph/interface/imu_raw" }, ... ] }, "config": { "carter_hardware.vlp16_initializer": { "lidar_initializer": { "pose": [1.0, 0.0, 0.0, 0.0, -0.04, 0.0, 0.59] } }, "scan_flattener.range_scan_flattening": { "isaac.perception.RangeScanFlattening": { "param": $(fullname carter_hardware.vlp16/lidar_initializer) } }, .... } }

2d_carter is itself a subgraph, contained in gmapping.app.json and various other sample applications included in the Isaac SDK.

Note that when creating an edge for IMU and when setting pose parameter for lidar, all the prefixes are added until the node definition is reached (only carter_hardware in this case). The complete node name in the application also depends on the name used when employing the 2d_carter subgraph, which is not known by this subgraph itself. To refer to the full name, the $(fullname <>) syntax can be used as illustrated above.

To prevent a node name being prefixed in the “edges” section, add a ‘/’ character to the edge name. In the example below, if the prefix due to the subgraph name is commander, the target is expanded to commander.virtual_gamepad_bridge/VirtualGamepadBridge/request, while the source reads websight/WebsightServer/virtual_gamepad thanks to the ‘/’ special character.

Copy
Copied!
            

"edges": [ { "source": "/websight/WebsightServer/virtual_gamepad", "target": "virtual_gamepad_bridge/VirtualGamepadBridge/request" }, ....

To make use of a subgraph in an application, it must be listed under the “data” parameter of the isaac_app Bazel function in the BUILD file. Declare the subgraph in the BUILD file by using the isaac_subgraph function.

Declare the subgraph with the isaac_subgraph function as shown below:

Copy
Copied!
            

load("//engine/build:isaac.bzl", "isaac_subgraph") isaac_subgraph( name = "carter_hardware_subgraph", subgraph = "carter_hardware.subgraph.json", modules = [ "imu", "segway", "velodyne_lidar" ], visibility = ["//visibility:public"], )

When using isaac_subgraph, the list of packages used by the subgraph can be enumerated and do not have to be repeated in the module list of an isaac_app that uses the subgraph.

To use the subgraph in an application, list it as a data parameter of an isaac_app as shown below:

Copy
Copied!
            

isaac_app( name = "carter", data = [ ... "//apps/carter:carter_hardware_subgraph", ], modules = [ ... ] )

To specify a prefix when loading a configuration file through the command line, use the following syntax:

Copy
Copied!
            

bob@desktop:~/isaac/sdk$ bazel run packages/freespace_dnn/apps:freespace_dnn_inference_image -- --config inference:packages/freespace_dnn/apps/freespace_dnn_inference_sidewalk_tensorrt.config.json


The inference: specification before the file name will result in the “inference” prefix being applied to all the nodes when loading packages/freespace_dnn/apps/freespace_dnn_inference_sidewalk_tensorrt.config.json.

Isaac SDK automatically has a global pose tree which can be used to compute the relative poses of 3D or 2D coordinate frames. The Isaac SDK pose tree also caches time histories of poses to allow queries relative to different time points.

If a component needs to read a pose it should use the ISAAC_POSE2 macro:

Copy
Copied!
            

// This is the pose under which the ego motion estimation will be written to the pose tree. ISAAC_POSE2(odom, robot) or :code:`ISAAC_POSE3` macro:

Copy
Copied!
            

// This provides access to the very same pose as above, in Pose3 format instead of Pose2. ISAAC_POSE3(odom, robot);

The ISAAC_POSE2 or ISAAC_POSE3 macro has two arguments which indicate the two coordinate frames in question. ISAAC_POSE2(lhs, rhs) will give the transformation lhs_T_rhs. This transformation can be used to transform a point in the rhs frame into a point in the lhs frame: p_lhs = lhs_T_rhs * p_rhs;.

In the case of DifferentialBaseOdometry the estimated pose of the robot relative to where it started is computed and written to the pose tree.

Copy
Copied!
            

const Pose2d odom_T_robot{SO2d::FromAngle(state_.heading()), Vector2d{state_.pos_x(), state_.pos_y()}}; set_odom_T_robot(odom_T_robot, getTickTime());

Copy
Copied!
            

// In case of using Pose3d, use the following line instead const Pose3d odom_T_robot{SO3d::FromAxisAngle(Vector3{0.0, 0.0, 1.0}, state_.heading()), Vector3d{state_.pos_x(), state_.pos_y(), state_.pos_z()}}; set_odom_T_robot(odom_T_robot, getTickTime());

Note that the function set_odom_T_robot (and similarly get_odom_T_robot) is generated automatically when the macro is used.

The pose is read with respect to a particular point in time. In this example getTickTime is used. Querying poses at various timepoints is crucial for time synchronization of data channels to avoid lag and data mismatch.

If you want to read a pose from the pose tree a similar mechanism could be used:

Copy
Copied!
            

const Pose2d foo_T_bar = get_foo_T_bar(getTickTime());

Copy
Copied!
            

// This is for Pose3d case const Pose3d foo_T_bar = get_foo_T_bar(getTickTime());

Pose3d provides 6-DOF information consisting 3-DOF orientation and 3-DOF translation thus works in general cases. Pose2d provides 1-DOF orientation and 2-DOF translation with assumption that the movement is constrained in X-Y plane with trivial pitch and roll, as most wheeled robots do. It is easier to work with and more efficient in such cases. Please choose one of them depending on the assumption for movement.

© Copyright 2018-2020, NVIDIA Corporation. Last updated on Oct 30, 2023.