WeRide's 2026 spring SDE OA reports on 1point3acres are clearly clustered around three tracks — perception task scheduling, graph node BFS, and path merging — covering ~90% of recent posts. This guide breaks down all three with Python solutions and shows how OA assist plugs in.
WeRide SDE OA 1point3acres Snapshot
| Dimension | Detail |
|---|---|
| Platform | Nowcoder + WeRide in-house assessment |
| Duration | 90 minutes |
| Question count | 2–3 |
| Focus | Perception / path / scheduling / graph theory |
| Grading | Auto-judged with hidden test data |
| Pass rate | ~42% per 1point3acres |
Track 1: Perception Task Scheduling
Problem
n sensor tasks each have (start, duration, priority). The agent runs one task at a time. Design a scheduling policy minimizing priority-weighted total completion time.
Python Solution (priority-aware SJF)
import heapq
def schedule_perception(tasks):
tasks.sort(key=lambda t: t[0])
heap = []
cur_time = 0
i = 0
n = len(tasks)
total = 0
while i < n or heap:
if not heap:
cur_time = max(cur_time, tasks[i][0])
while i < n and tasks[i][0] <= cur_time:
s, d, p = tasks[i]
heapq.heappush(heap, (d / p, d, p))
i += 1
_, d, p = heapq.heappop(heap)
cur_time += d
total += cur_time * p
return total
Complexity: O(n log n). Hidden cases on 1point3acres include priority=0 and fully overlapping windows.
Track 2: Graph Node BFS (Perception Graph Traversal)
Problem
Given an undirected graph (vehicle perception output), find every node reachable from start within k hops. The returned path must follow lexicographic order.
Python Solution
from collections import defaultdict, deque
def k_hop_neighbors(edges, start, k):
g = defaultdict(list)
for u, v in edges:
g[u].append(v)
g[v].append(u)
for u in g:
g[u].sort()
visited = {start: 0}
queue = deque([(start, 0)])
result = []
while queue:
node, d = queue.popleft()
if d > 0:
result.append(node)
if d == k:
continue
for nb in g[node]:
if nb not in visited:
visited[nb] = d + 1
queue.append((nb, d + 1))
return result
Complexity: O(V + E). Common stumbles: self-loop (u,u), duplicate edges, isolated nodes.
Track 3: Path Merge
Problem
Vehicle has n planned path segments (start_x, start_y, end_x, end_y). Merge all head-to-tail connected segments into one path, ordered by departure time.
Python Solution
def merge_path(segments):
nxt = {}
starts = set()
ends = set()
for s in segments:
nxt[(s[0], s[1])] = (s[2], s[3])
starts.add((s[0], s[1]))
ends.add((s[2], s[3]))
head_candidates = starts - ends
if len(head_candidates) != 1:
return []
head = head_candidates.pop()
path = [head]
while head in nxt:
head = nxt[head]
path.append(head)
return path
Complexity: O(n). Watch for "disconnected" and "cyclic" corner cases — 1point3acres reports at least one cycle-detection trap.
1point3acres Frequency Cheat Sheet
| Track | 60-day Frequency | Algorithm | Common Bug |
|---|---|---|---|
| Perception scheduling | ★★★★★ | Priority queue | priority=0 |
| K-hop BFS | ★★★★ | BFS + lex order | Self-loop / dup |
| Path merge | ★★★★ | Adjacency + head | Cycle |
| Grid fill | ★★★ | DFS / Union-Find | Out-of-bound |
| Quote parsing | ★★ | State machine | Missing field |
OA Assist Playbook
What oavoservice OA assist gives you
- Question bucketing: 30-day WeRide posts on 1point3acres bucketed into perception / graph / path with 4 variants each
- Timed simulation: 90-minute full-pace run with mentor
- Live OA assist: low-latency idea check on the in-house platform
- VO transition: same mentor carries into onsite VO covering system design + BQ + AV business questions
Beyond 1point3acres
Question wordings shift, but the skeleton (perception + graph + path) is stable. We maintain a L4 AV terminology + high-frequency interview scenario glossary so you can solve the algorithm and correctly parse "Obstacle / Trajectory / World Frame" wording in the same 90 minutes.
Add WeChat Coding0201 for pricing and scope.
FAQ
Are SDE OA and SDE-Perception OA the same bank?
Not exactly. General SDE leans graph / path; Perception adds 1 simulation question on point clouds / trajectories.
Why does 1point3acres report ~42% pass rate?
The two killers are perception scheduling hidden cases and path merge cycle detection. We've watched submissions stuck at 9/13 on a priority=0 hidden case.
What languages does WeRide allow?
Python / C++ / Java; ~65% of 1point3acres reports use Python.
Cooldown after a fail?
WeRide 6 months. Cross-BU (AV / mobility / freight) typically uses a separate pool.
Preparing for WeRide / Pony.ai / Waymo OA / VO?
oavoservice tracks the AV space (WeRide / Pony.ai / Waymo / Cruise / Zoox) end-to-end. Our mentors come from perception / decision / simulation teams and deliver question bucketing, timed simulation, live OA assist on the in-house platform, and AV terminology training.
👉 Add WeChat: Coding0201 for the WeRide 1point3acres question bank and OA assist plan.
Contact
Email: [email protected]
Telegram: @OAVOProxy