added move base based on distance/degrees

Author: hello-vinitha

ai-gateway/constants.d.ts

@@ -1,6 +1,10 @@
 export declare const VOICE_DURATION_MS_MIN: 100;
 export declare const VOICE_DURATION_MS_MAX: 10000;
 export declare const VOICE_DURATION_MS_DEFAULT: 800;
+export declare const VOICE_DISTANCE_M_MIN: number;
+export declare const VOICE_DISTANCE_M_MAX: number;
+export declare const VOICE_ROTATION_DEG_MIN: number;
+export declare const VOICE_ROTATION_DEG_MAX: number;
 export declare const EXECUTE_BASE_MOVE: "execute_base_move";
 export declare const STOP_BASE_MOVE: "stop_base_move";
 export declare const REPEAT_BASE_MOVE: "repeat_base_move";

ai-gateway/constants.js

@@ -13,6 +13,21 @@ const VOICE_DURATION_MS_MAX = 10000;
 /** Server schema default when the model omits duration_ms. */
 const VOICE_DURATION_MS_DEFAULT = 800;
 
+/**
+ * Minimum distance (meters) for distance-based translation moves.
+ * Rotation distance_m is in degrees; this bound is reused for both.
+ */
+const VOICE_DISTANCE_M_MIN = 0.05;
+
+/** Maximum distance (meters) for distance-based translation; or degrees for rotation. */
+const VOICE_DISTANCE_M_MAX = 5.0;
+
+/** Minimum rotation (degrees) for distance-based rotation. */
+const VOICE_ROTATION_DEG_MIN = 1.0;
+
+/** Maximum rotation (degrees) for distance-based rotation. */
+const VOICE_ROTATION_DEG_MAX = 360.0;
+
 /** Timed holonomic base translation or in-place rotation (action, speed, duration_ms). */
 const EXECUTE_BASE_MOVE = "execute_base_move";
 
@@ -52,6 +67,10 @@ module.exports = {
     VOICE_DURATION_MS_MIN,
     VOICE_DURATION_MS_MAX,
     VOICE_DURATION_MS_DEFAULT,
+    VOICE_DISTANCE_M_MIN,
+    VOICE_DISTANCE_M_MAX,
+    VOICE_ROTATION_DEG_MIN,
+    VOICE_ROTATION_DEG_MAX,
     EXECUTE_BASE_MOVE,
     STOP_BASE_MOVE,
     REPEAT_BASE_MOVE,

ai-gateway/openai-realtime-api.js

@@ -21,6 +21,10 @@ const {
     VOICE_DURATION_MS_DEFAULT,
     VOICE_DURATION_MS_MAX,
     VOICE_DURATION_MS_MIN,
+    VOICE_DISTANCE_M_MIN,
+    VOICE_DISTANCE_M_MAX,
+    VOICE_ROTATION_DEG_MIN,
+    VOICE_ROTATION_DEG_MAX,
 } = require("./constants");
 
 /**
@@ -67,8 +71,11 @@ function buildRealtimeVoiceSessionPayload() {
                 "If ambiguous, prefer slower/shorter movements.",
                 `Call tool ${EXECUTE_BASE_MOVE} at most ONCE per user utterance.`,
                 `After a tool result, confirm warmly in one brief sentence — never call ${EXECUTE_BASE_MOVE} again unless the user gives a new command.`,
-                `Call tool ${EXECUTE_BASE_MOVE} with action (enum), speed (${BASE_MOVE_SPEEDS.join("|")}), duration_ms (${VOICE_DURATION_MS_MIN}-${VOICE_DURATION_MS_MAX}).`,
-                `Default duration_ms ${VOICE_DURATION_MS_DEFAULT} if unstated; default speed ${BASE_MOVE_SPEED_DEFAULT}.`,
+                `Call tool ${EXECUTE_BASE_MOVE} with action (enum), speed (${BASE_MOVE_SPEEDS.join("|")}), and EITHER duration_ms OR distance_m — not both.`,
+                `If the user specifies a distance (e.g. "move forward half a meter", "go forward 0.5 meters"), set distance_m in meters (${VOICE_DISTANCE_M_MIN}–${VOICE_DISTANCE_M_MAX}) and omit duration_ms.`,
+                `If the user specifies a rotation angle (e.g. "rotate 90 degrees", "turn left a quarter turn"), set distance_m in degrees (e.g. 90) and omit duration_ms.`,
+                `If the user specifies a time (e.g. "move forward for 2 seconds"), set duration_ms and omit distance_m.`,
+                `If neither distance nor time is specified, use default duration_ms ${VOICE_DURATION_MS_DEFAULT} and omit distance_m.`,
                 `When the user wants to halt base motion (stop, wait, freeze, do not move, cut that, enough), call tool ${STOP_BASE_MOVE} with no arguments — not ${EXECUTE_BASE_MOVE}.`,
                 `When the user wants to repeat the last base move (again, same thing, one more time, do that again), call tool ${REPEAT_BASE_MOVE} with no arguments — not ${EXECUTE_BASE_MOVE} with guessed parameters. Ensure you are over 90% confident that the user asked you to repeat previous movement before you move.`,
                 "Do not respond to the user. Do not speak to the user. Only call tools as instructed.",
@@ -97,11 +104,21 @@ function buildRealtimeVoiceSessionPayload() {
                             duration_ms: {
                                 type: "integer",
                                 description:
-                                    "How long to move in milliseconds before stopping.",
+                                    "How long to move in milliseconds before stopping. Use when the user specifies a time. Omit if distance_m is set.",
                                 minimum: VOICE_DURATION_MS_MIN,
                                 maximum: VOICE_DURATION_MS_MAX,
                                 default: VOICE_DURATION_MS_DEFAULT,
                             },
+                            distance_m: {
+                                type: "number",
+                                description:
+                                    "Distance to travel in meters (for translation actions: forward, backward, strafe). " +
+                                    "For rotation actions (rotate_left, rotate_right), pass the angle in degrees (e.g. 90 for a quarter turn). " +
+                                    `Valid range: translation (${VOICE_DISTANCE_M_MIN}–${VOICE_DISTANCE_M_MAX} meters), ` +
+                                    `rotation (${VOICE_ROTATION_DEG_MIN}–${VOICE_ROTATION_DEG_MAX} degrees). Omit if duration_ms is set.`,
+                                minimum: VOICE_DISTANCE_M_MIN,
+                                maximum: VOICE_ROTATION_DEG_MAX,
+                            },
                         },
                         required: ["action"],
                         additionalProperties: false,

src/pages/operator/tsx/voice/constants.ts

@@ -15,6 +15,10 @@ import {
     VOICE_DURATION_MS_DEFAULT,
     VOICE_DURATION_MS_MAX,
     VOICE_DURATION_MS_MIN,
+    VOICE_DISTANCE_M_MIN,
+    VOICE_DISTANCE_M_MAX,
+    VOICE_ROTATION_DEG_MIN,
+    VOICE_ROTATION_DEG_MAX,
 } from "ai-gateway/constants";
 
 /**
@@ -32,6 +36,10 @@ export {
     VOICE_DURATION_MS_DEFAULT,
     VOICE_DURATION_MS_MAX,
     VOICE_DURATION_MS_MIN,
+    VOICE_DISTANCE_M_MIN,
+    VOICE_DISTANCE_M_MAX,
+    VOICE_ROTATION_DEG_MIN,
+    VOICE_ROTATION_DEG_MAX,
 };
 
 /** Derived from ai-gateway action/speed enums (operator client). */
@@ -48,6 +56,14 @@ export type ExecuteBaseMoveArgs = {
     action: BaseMoveAction;
     speed?: BaseMoveSpeed;
     duration_ms?: number;
+    /**
+     * Distance-based move target.
+     * - Translation actions (forward/backward/strafe): meters.
+     * - Rotation actions (rotate_left/rotate_right): degrees (converted to radians client-side).
+     * When set, the robot moves until odom confirms the target is reached instead of relying
+     * on a fixed time duration. Falls back to an estimated duration if odom is unavailable.
+     */
+    distance_m?: number;
 };
 
 /** Tool result sent back on the Realtime data channel (operator client). */
@@ -111,6 +127,28 @@ export function clampDurationMs(ms: number): number {
     );
 }
 
+/**
+ * Clamp a distance_m value from tool args before robot dispatch.
+ * - For translation: value is in meters.
+ * - For rotation: value is in degrees (caller converts to radians).
+ *
+ * @param d raw distance from Realtime tool arguments
+ * @returns clamped value in [VOICE_DISTANCE_M_MIN, VOICE_DISTANCE_M_MAX]
+ */
+export function clampDistanceM(d: number): number {
+    return Math.max(VOICE_DISTANCE_M_MIN, Math.min(VOICE_DISTANCE_M_MAX, d));
+}
+
+/**
+ * Clamp a rotation degrees value from tool args before robot dispatch.
+ *
+ * @param d raw rotation in degrees from Realtime tool arguments
+ * @returns clamped value in [VOICE_ROTATION_DEG_MIN, VOICE_ROTATION_DEG_MAX]
+ */
+export function clampRotationDeg(d: number): number {
+    return Math.max(VOICE_ROTATION_DEG_MIN, Math.min(VOICE_ROTATION_DEG_MAX, d));
+}
+
 /** Union of tool names in VOICE_TOOLS; used for typed dispatch in realtimeSession.ts. */
 export type VoiceToolName = (typeof VOICE_TOOLS)[number];
 

src/pages/operator/tsx/voice/executeBaseMove.ts

@@ -8,7 +8,9 @@ import {
     BASE_MOVE_ACTIONS,
     BASE_MOVE_SPEED_DEFAULT,
     BASE_MOVE_SPEEDS,
+    clampDistanceM,
     clampDurationMs,
+    clampRotationDeg,
     VOICE_DURATION_MS_DEFAULT,
     VOICE_MOVE_DEDUPE_MS,
     type BaseMoveAction,
@@ -56,6 +58,7 @@ let lastExecutedMove: {
     action: BaseMoveAction;
     speed: BaseMoveSpeed;
     duration_ms: number;
+    distance_m: number | undefined;
     at: number;
 } | null = null;
 
@@ -83,6 +86,7 @@ function isDuplicateVoiceMove(
     action: BaseMoveAction,
     speed: BaseMoveSpeed,
     duration_ms: number,
+    distance_m: number | undefined,
 ): boolean {
     if (lastExecutedMove === null) {
         return false;
@@ -94,7 +98,8 @@ function isDuplicateVoiceMove(
     return (
         lastExecutedMove.action === action &&
         lastExecutedMove.speed === speed &&
-        lastExecutedMove.duration_ms === duration_ms
+        lastExecutedMove.duration_ms === duration_ms &&
+        lastExecutedMove.distance_m === distance_m
     );
 }
 
@@ -139,6 +144,7 @@ function formatMoveOkDetail(
     action: BaseMoveAction,
     speed: BaseMoveSpeed,
     duration_ms: number,
+    distance_m: number | undefined,
     mode: VoiceMoveExecutionMode,
     linX: number,
     linY: number,
@@ -149,11 +155,14 @@ function formatMoveOkDetail(
         mode === "button_provider"
             ? ` via ${button ?? "button"}`
             : " via direct drive";
-    let detail = `${action} at ${speed} for ${duration_ms}ms${via}`;
+    const distanceStr = distance_m !== undefined
+        ? ` for ${distance_m.toFixed(5)} m (odom-tracked)`
+        : ` for ${duration_ms}ms`;
+    let detail = `${action} at ${speed}${distanceStr}${via}`;
     if (mode === "direct") {
-        detail += ` (~linVelX=${linX.toFixed(4)}, linVelY=${linY.toFixed(4)}`;
+        detail += ` (~linVelX=${linX.toFixed(5)}, linVelY=${linY.toFixed(5)}`;
         if (angVel !== 0) {
-            detail += `, angVel=${angVel.toFixed(4)}`;
+            detail += `, angVel=${angVel.toFixed(5)}`;
         }
         detail += ")";
     }
@@ -183,6 +192,7 @@ export function coerceExecuteArgs(raw: Record<string, unknown>): {
     action: BaseMoveAction;
     speed: BaseMoveSpeed;
     duration_ms: number;
+    distance_m: number | undefined;
 } | null {
     const action = raw.action as string | undefined;
     if (!action || !VALID_EXECUTE_ACTIONS.has(action)) {
@@ -195,6 +205,22 @@ export function coerceExecuteArgs(raw: Record<string, unknown>): {
     }
     const speed = speedRaw as BaseMoveSpeed;
 
+    const isRotation = action === "rotate_left" || action === "rotate_right";
+
+    // Parse distance_m first — if present it takes priority over duration_ms.
+    let distance_m: number | undefined;
+    if (raw.distance_m !== undefined && raw.distance_m !== null) {
+        const rawDist =
+            typeof raw.distance_m === "number"
+                ? raw.distance_m
+                : typeof raw.distance_m === "string"
+                    ? Number.parseFloat(raw.distance_m)
+                    : NaN;
+        if (!Number.isNaN(rawDist) && rawDist > 0) {
+            distance_m = isRotation ? clampRotationDeg(rawDist) : clampDistanceM(rawDist);
+        }
+    }
+
     let duration_ms =
         typeof raw.duration_ms === "number"
             ? raw.duration_ms
@@ -208,7 +234,7 @@ export function coerceExecuteArgs(raw: Record<string, unknown>): {
 
     duration_ms = clampDurationMs(duration_ms);
 
-    return { action: action as BaseMoveAction, speed, duration_ms };
+    return { action: action as BaseMoveAction, speed, duration_ms, distance_m };
 }
 
 export function executeBaseMoveOnProvider(
@@ -225,11 +251,11 @@ export function executeBaseMoveOnProvider(
         };
     }
 
-    const { action, speed, duration_ms } = coerced;
+    const { action, speed, duration_ms, distance_m } = coerced;
 
     if (
         !opts?.skipDedupe &&
-        isDuplicateVoiceMove(action, speed, duration_ms)
+        isDuplicateVoiceMove(action, speed, duration_ms, distance_m)
     ) {
         return {
             ok: false,
@@ -241,6 +267,50 @@ export function executeBaseMoveOnProvider(
     const mode = voiceMoveExecutionContext?.mode ?? "direct";
     prepareVoiceMoveUi(speed);
 
+    // --- Distance-based path ---
+    // Compute an estimated duration from distance / velocity, then dispatch
+    // through the standard timed path. Safe, simple, no odom dependency.
+    if (distance_m !== undefined) {
+        const isRotation = action === "rotate_left" || action === "rotate_right";
+        const { linX, linY, angVel } = velocitiesForAction(action);
+
+        // For rotation the model sends degrees; convert to radians for time estimation.
+        const targetNative = isRotation ? (distance_m * Math.PI) / 180 : distance_m;
+        const speed_mps = isRotation
+            ? Math.abs(angVel)
+            : Math.max(Math.abs(linX), Math.abs(linY));
+        const estimatedMs = speed_mps > 0
+            ? Math.round((targetNative / speed_mps) * 1000)
+            : VOICE_DURATION_MS_DEFAULT;
+        const clampedMs = clampDurationMs(estimatedMs);
+
+        if (mode === "button_provider") {
+            const button = VOICE_ACTION_TO_BUTTON[action];
+            const started = provider.timedButtonPadPress(button, clampedMs);
+            if (!started) {
+                return busyOrDisconnectedResult(provider, "timedButtonPadPress");
+            }
+            lastVoiceBaseMove = { action, speed, duration_ms: clampedMs, distance_m };
+            lastExecutedMove = { action, speed, duration_ms: clampedMs, distance_m, at: Date.now() };
+            return {
+                ok: true,
+                detail: formatMoveOkDetail(action, speed, clampedMs, distance_m, mode, 0, 0, 0, button),
+            };
+        }
+
+        const started = provider.timedBaseDrive(linX, linY, clampedMs, angVel);
+        if (!started) {
+            return busyOrDisconnectedResult(provider, "timedBaseDrive");
+        }
+        lastVoiceBaseMove = { action, speed, duration_ms: clampedMs, distance_m };
+        lastExecutedMove = { action, speed, duration_ms: clampedMs, distance_m, at: Date.now() };
+        return {
+            ok: true,
+            detail: formatMoveOkDetail(action, speed, clampedMs, distance_m, mode, linX, linY, angVel),
+        };
+    }
+
+    // --- Duration-based path (original) ---
     if (mode === "button_provider") {
         const button = VOICE_ACTION_TO_BUTTON[action];
         const started = provider.timedButtonPadPress(button, duration_ms);
@@ -251,13 +321,14 @@ export function executeBaseMoveOnProvider(
             );
         }
         lastVoiceBaseMove = { action, speed, duration_ms };
-        lastExecutedMove = { action, speed, duration_ms, at: Date.now() };
+        lastExecutedMove = { action, speed, duration_ms, distance_m: undefined, at: Date.now() };
         return {
             ok: true,
             detail: formatMoveOkDetail(
                 action,
                 speed,
                 duration_ms,
+                undefined,
                 mode,
                 0,
                 0,
@@ -274,14 +345,15 @@ export function executeBaseMoveOnProvider(
     }
 
     lastVoiceBaseMove = { action, speed, duration_ms };
-    lastExecutedMove = { action, speed, duration_ms, at: Date.now() };
+    lastExecutedMove = { action, speed, duration_ms, distance_m: undefined, at: Date.now() };
 
     return {
         ok: true,
         detail: formatMoveOkDetail(
             action,
             speed,
             duration_ms,
+            undefined,
             mode,
             linX,
             linY,

src/pages/operator/tsx/voice/realtimeSession.ts

@@ -87,11 +87,16 @@ function formatToolCallLog(
                 string,
                 unknown
             >;
-            argsSummary = JSON.stringify({
+            const summary: Record<string, unknown> = {
                 action: parsed.action,
                 speed: parsed.speed ?? BASE_MOVE_SPEED_DEFAULT,
-                duration_ms: parsed.duration_ms ?? VOICE_DURATION_MS_DEFAULT,
-            });
+            };
+            if (parsed.distance_m !== undefined && parsed.distance_m !== null) {
+                summary.distance_m = parsed.distance_m;
+            } else {
+                summary.duration_ms = parsed.duration_ms ?? VOICE_DURATION_MS_DEFAULT;
+            }
+            argsSummary = JSON.stringify(summary);
         } catch {
             //
         }