05. Position Control

Drives the Inverse3 cursor toward a target position via set_cursor_position. The interaction model differs by language — C++ uses one-shot random targets, Python uses continuous keyboard-driven motion.

What you'll learn:

• Using set_cursor_position for position-mode control
• Two different interaction models for the same underlying command
• Clamping the target to a workspace sphere — Minverse uses a smaller radius than Inverse3
• Setting a workspace preset so the origin sits at workspace centre

Workflow

C++ (random-target model)

1. Start a background input thread that reads line-buffered keystrokes (n, +, -, q) from stdin.
2. Open the WebSocket. On the first state frame, register the session profile and set configure.preset: arm_front_centered. Generate the first random target inside a sphere (rejection sampling, radius 0.08 m).
3. Each tick, send a set_cursor_position command to the current target. The cursor smoothly tracks it — the service rate-limits and interpolates.
4. When the user types n + ENTER, the input thread generates a new random target. + / - adjust speed; q quits.

Python (hold-to-move model)

1. Open the WebSocket. On the first state frame, check status.calibrated — prompt the user if the device isn't calibrated yet.
2. Read config.type to choose the workspace radius (minverse = 0.04 m, anything else = 0.10 m).
3. Register the session profile and set configure.preset: arm_front_centered.
4. Each tick: poll keyboard state (W/A/S/D/Q/E), update the target position by SPEED along each pressed axis, clamp to the workspace sphere, and send set_cursor_position. R resets the target to the origin.

Parameters

Name	Default (C++)	Default (Python)	Purpose
workspace_radius / RADIUS_INVERSE3	0.08 m	0.10 m (Inverse3) / 0.04 m (Minverse)	Target sphere radius
speed_step / SPEED	0.01 / press	0.00005 m / tick	Step per interaction
PRINT_EVERY_MS	—	100	Telemetry throttle (Python)
Session profile	co.haply.inverse.tutorials:position-control	same	Identifies in Haply Hub

Calibration check (Python)

The Python variant checks status.calibrated from the first state frame and prompts the user if the device isn't calibrated. The C++ variant assumes calibration is already complete.

State fields read

• data.inverse3[0].device_id — for building the command
• data.inverse3[0].state.cursor_position — telemetry
• (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

Communication workflow

• C++ runs a background stdin thread that writes std::atomic<float> targets; the WebSocket thread reads them each tick. On n + ENTER, the input thread generates a new random target; on q, both threads shut down.
• Python is single-threaded async — the WebSocket loop polls keyboard state each tick and updates position directly.

The Inverse-API payloads are the same: first tick carries the session profile + configure.preset, subsequent ticks carry only set_cursor_position.

Python

Single async loop. Keyboard polling (handle_keys) runs inline every tick — no threads. config.type and status.calibrated are read once from the first state frame.

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

        if first_message:
            first_message = False
            device_id = data["inverse3"][0]["device_id"]
            radius    = get_workspace_radius(data["inverse3"][0].get("config", {}))
            # Handshake: profile + preset (one-shot)
            request_msg = {
                "session": {"configure": {"profile": {"name": SLUG}}},
                "inverse3": [{
                    "device_id": device_id,
                    "configure": {"preset": {"preset": "arm_front_centered"}},
                }],
            }
        else:
            # Per tick: update position from keyboard (classic polling, not shown), send command
            position = handle_keys(position, radius)
            request_msg = {
                "inverse3": [{
                    "device_id": device_id,
                    "commands": {"set_cursor_position": {"position": position}},
                }],
            }

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

C++ (nlohmann)

Two-thread model: a background thread reads stdin and writes std::atomic<float> targets; the libhv I/O thread runs ws.onmessage each tick and reads the atomics.

// Shared state written by the stdin thread, read by the ws thread
static std::atomic<float> target_x{0.0f}, target_y{0.0f}, target_z{0.0f};

ws.onmessage = [&](const std::string &message) {
    const json data = json::parse(message);
    if (data["inverse3"].empty()) return;

    json request = {};
    const bool do_handshake = first_message.exchange(false);
    if (do_handshake) {
        request["session"] = {{"configure", {{"profile",
            {{"name", "co.haply.inverse.tutorials:position-control"}}}}}};
        generate_random_target();
    }

    request["inverse3"] = json::array();
    for (auto &el : data["inverse3"].items()) {
        json dev = {
            {"device_id", el.value()["device_id"]},
            {"commands", {{"set_cursor_position",
                {{"position", {{"x", target_x.load()},
                               {"y", target_y.load()},
                               {"z", target_z.load()}}}}}}},
        };
        if (do_handshake)
            dev["configure"] = {{"preset", {{"preset", "arm_front_centered"}}}};
        request["inverse3"].push_back(dev);
    }
    ws.send(request.dump());
};
ws.open("ws://localhost:10001");
std::thread input_thr(input_thread_func);  // stdin reader — writes target atomics
while (running) std::this_thread::sleep_for(100ms);

C++ (Glaze)

Same two-thread model. Typed structs for the command — std::optional<device_configure> carries the one-shot preset per-device; omitted from JSON on subsequent ticks.

// Struct models
struct vec3 { float x{}, y{}, z{}; };
struct set_cursor_position_cmd { vec3 position; };
struct preset_cfg { std::string preset; };
struct device_configure { std::optional<preset_cfg> preset; };

struct device_commands {
    std::string device_id;
    std::optional<device_configure> configure;  // one-shot
    struct commands_t {
        std::optional<set_cursor_position_cmd> set_cursor_position;
    } commands;
};
struct commands_message {
    std::optional<session_cmd> session;         // one-shot
    std::vector<device_commands> inverse3;
};

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

    commands_message request;
    const bool do_handshake = first_message.exchange(false);
    if (do_handshake) {
        request.session = session_cmd{ /* profile = position-control */ };
    }

    for (const auto &dev : data.inverse3) {
        device_commands dc{ .device_id = dev.device_id };
        dc.commands.set_cursor_position = set_cursor_position_cmd{
            .position = {target_x.load(), target_y.load(), target_z.load()}};
        if (do_handshake)
            dc.configure = device_configure{ .preset = preset_cfg{"arm_front_centered"} };
        request.inverse3.push_back(std::move(dc));
    }

    std::string out;
    (void)glz::write_json(request, out);
    ws.send(out);
};
ws.open("ws://localhost:10001");
std::thread input_thr(input_thread_func);
while (running) std::this_thread::sleep_for(100ms);

Source: Python · C++ · C++ Glaze

Related: Control Commands (set_cursor_position) · Mount & Workspace (presets) · Types (vec3) · Tutorial 06 (Combined)