06. Combined (Inverse3 + Wireless VerseGrip)

Two-device tutorial: point the grip and hold a button to drive the Inverse3 cursor in that direction. The cursor is clamped inside a spherical workspace.

What you'll learn:

• Reading two device types in the same state frame (inverse3 and wireless_verse_grip)
• Extracting the grip's pointing direction from its quaternion (local +Y axis)
• Using set_cursor_position to drive the cursor toward a computed target
• Clamping the target to a safe workspace sphere — Minverse uses a smaller radius than Inverse3
• Setting a workspace preset (arm_front_centered) so the origin sits at the middle of reach

Workflow

1. Discover both devices:
   • C++ variants query GET /devices over HTTP at startup, then print a calibration prompt and wait for ENTER.
   • Python reads both device IDs from the first WebSocket state frame.
2. Register the session profile and set configure.preset: arm_front_centered on the first message (one-shot handshake).
3. Each tick: read the grip's orientation and buttons.{a, b} state.
4. If a motion button is held, compute the grip's world-space direction (R(q) · ĵ — the rotated unit +Y axis) and accumulate it into the target position scaled by SPEED.
5. Clamp the target inside the workspace sphere and send it via set_cursor_position.
6. (Python only) Adapt the sphere radius from the device's config.type — minverse = 0.04 m, everything else = 0.10 m.

Parameters

Name	Default	Purpose
SPEED	0.01 m/tick	Motion step while a button is held
RADIUS_INVERSE3	0.10 m	Workspace clamp radius for Inverse3 / Inverse3x
RADIUS_MINVERSE	0.04 m	Workspace clamp radius for Minverse (Python only — C++ hardcodes 0.10)
PRINT_EVERY_MS	200	Telemetry throttle
Session profile name	co.haply.inverse.tutorials:combined	Identifies this simulation in Haply Hub

Calibrate before running

• Let the Inverse3 self-calibrate (or place the grip on the inkwell and wait for the LED to turn solid).
• Detach the grip from the inkwell.
• Hold A or B and rotate the grip — the cursor moves along the direction the grip is pointing.

State fields read

From the per-tick state frame:

• data.inverse3[0].state.cursor_position — vec3
• data.wireless_verse_grip[0].state.orientation — quaternion
• data.wireless_verse_grip[0].state.buttons.{a, b, c} — booleans
• (Python, first frame only) data.inverse3[0].config.type — selects Inverse3 vs Minverse radius
• (Python, first frame only) data.inverse3[0].status.calibrated — prompts the user if false

Send / receive

The quaternion-to-direction math (rotate +Y by R(q)) and sphere clamp are classic linear algebra — see the source files. The Inverse-API side is the handshake + per-tick set_cursor_position.

Python

Single async loop. Python reads both device IDs from the first state frame; the handshake attaches profile + configure.preset: arm_front_centered to the first set_cursor_position.

async with websockets.connect(URI) as websocket:
    while True:
        msg  = await websocket.recv()
        data = json.loads(msg)

        if first_message:
            first_message = False
            inverse3_id = data["inverse3"][0]["device_id"]
            grip_id     = data["wireless_verse_grip"][0]["device_id"]
            radius      = get_workspace_radius(data["inverse3"][0].get("config", {}))

            # Handshake: profile + preset + first position command
            request_msg = {
                "session": {"configure": {"profile": {"name": SLUG}}},
                "inverse3": [{
                    "device_id": inverse3_id,
                    "configure": {"preset": {"preset": "arm_front_centered"}},
                    "commands": {"set_cursor_position": {"position": position}},
                }],
            }
        else:
            # Per tick: update position from grip pointing direction (classic math, not shown), send
            request_msg = {
                "inverse3": [{
                    "device_id": inverse3_id,
                    "commands": {"set_cursor_position": {"position": position}},
                }],
            }

        await websocket.send(json.dumps(request_msg))

C++ (nlohmann)

C++ discovers both device IDs via GET /devices at startup (HTTP), then opens the WebSocket. onmessage runs on libhv's I/O thread; the main thread blocks on ENTER.

// Startup (synchronous):
const std::string inv3_device_id = get_first_device_id("inverse3");
const std::string grip_device_id = get_first_device_id("wireless_verse_grip");

// Per tick:
ws.onmessage = [&](const std::string &msg) {
    json data = json::parse(msg);
    // ... classic math (not shown): update local Inverse3State + WirelessVerseGripState,
    //     compute new position from grip orientation, clamp to sphere ...

    json command;
    command["inverse3"] = json::array();
    command["inverse3"].push_back({
        {"device_id", inv3_device_id},
        {"commands", {{"set_cursor_position",
            {{"position", {{"x", pos.x}, {"y", pos.y}, {"z", pos.z}}}}}}}
    });

    if (first_message) {
        first_message = false;
        command["session"] = {{"configure", {{"profile",
            {{"name", "co.haply.inverse.tutorials:combined"}}}}}};
        command["inverse3"][0]["configure"] = {
            {"preset", {{"preset", "arm_front_centered"}}}};
    }

    ws.send(command.dump());
};
ws.open("ws://localhost:10001");
while (std::cin.get() != '\n') {}  // block main thread

C++ (Glaze)

Same libhv callback model. Typed structs model both state frames and the outgoing command — two std::vector<> in devices_message let a single glz::read yield both device types.

// Struct models
struct quat { float w{1.0f}, x{}, y{}, z{}; };
struct button_state { bool a{}, b{}, c{}; };
struct wvg_state { quat orientation{}; uint8_t hall{}; button_state buttons{}; };
struct wvg_device { std::string device_id; wvg_state state; };

struct inverse_state { vec3 cursor_position{}, cursor_velocity{}; };
struct inverse_device { std::string device_id; inverse_state state; };

struct devices_message {
    std::vector<inverse_device> inverse3;
    std::vector<wvg_device> wireless_verse_grip;
};

struct set_cursor_position_cmd { vec3 position; };
struct commands_message {
    std::optional<session_cmd> session;
    std::vector<device_commands> inverse3;
};

// Send / receive
ws.onmessage = [&](const std::string &msg) {
    devices_message data{};
    if (glz::read<glz_settings>(data, msg)) return;

    const auto &wvg = data.wireless_verse_grip[0].state;
    cursor_pos = data.inverse3[0].state.cursor_position;
    // ... classic math (not shown): if (wvg.buttons.a || wvg.buttons.b)
    //     move in pointing dir; clamp to sphere ...

    commands_message out_cmds{};
    device_commands dc{ .device_id = inv3_device_id };
    dc.commands.set_cursor_position = set_cursor_position_cmd{cursor_pos};

    if (first_message) {
        first_message = false;
        out_cmds.session = session_cmd{ /* profile = combined */ };
        dc.configure = device_configure{ .preset = preset_cfg{"arm_front_centered"} };
    }
    out_cmds.inverse3.push_back(std::move(dc));

    std::string out_json;
    (void)glz::write_json(out_cmds, out_json);
    ws.send(out_json);
};
ws.open("ws://localhost:10001");
while (std::cin.get() != '\n') {}  // block main thread

Source: Python · C++ · C++ Glaze

Related: Control Commands (set_cursor_position) · Types (quaternion, vec3) · Mount & Workspace (presets) · Tutorial 03 (Wireless VG) · Tutorial 05 (Position Control)